Technical Review: Architecting an AI-Driven Decision Support System for Enhanced Online Learning and Assessment

Abstract

The rapid expansion of online learning platforms has necessitated advanced systems to address scalability, personalization, and assessment challenges. This paper presents a comprehensive review of artificial intelligence (AI)-based decision support systems (DSSs) designed for online learning and assessment, synthesizing advancements from 2020 to 2025. By integrating machine learning, natural language processing, knowledge-based systems, and deep learning, AI-DSSs enhance educational outcomes through predictive analytics, automated grading, and personalized learning paths. This study examines system architecture, data requirements, model selection, and user-centric design, emphasizing their roles in achieving scalability and inclusivity. Through case studies of a MOOC platform using NLP and an adaptive learning system employing reinforcement learning, this paper highlights significant improvements in grading efficiency (up to 70%) and student performance (12–20% grade increases). Performance metrics, including accuracy, response time, and user satisfaction, are analyzed alongside evaluation frameworks combining quantitative and qualitative approaches. Technical challenges, such as model interpretability and bias, ethical concerns like data privacy, and implementation barriers, including cost and adoption resistance, are critically assessed, with proposed mitigation strategies. Future directions explore generative AI, multimodal integration, and cross-cultural studies to enhance global accessibility. This review offers a robust framework for researchers and practitioners, providing actionable insights for designing equitable, efficient, and scalable AI-DSSs to transform online education.

1. Introduction

The rapid expansion of online learning platforms over the past decade has fundamentally reshaped educational paradigms, enabling equitable access to knowledge across diverse global populations. Since 2020, the adoption of e-learning has accelerated, driven by technological advancements and catalyzed by the COVID-19 pandemic, which necessitated a swift transition to digital education [1]. By 2025, platforms such as Coursera, edX, and institution-specific learning management systems (LMSs) have reported cumulative enrollments exceeding 500 million users, underscoring the scalability and flexibility of online learning [2,3]. Despite these advances, traditional online learning systems face persistent challenges, including scalability limitations, inadequate personalization, and inefficiencies in delivering real-time, actionable feedback [4]. These shortcomings constrain the ability to address diverse learner needs and optimize educational outcomes effectively.

AI-driven decision support systems have emerged as a transformative approach to overcoming these challenges. By integrating machine learning (ML), natural language processing (NLP), and knowledge-based systems, AI-DSSs enable dynamic, data-driven decision-making in online learning environments. These systems facilitate personalized learning pathways, automate assessments, and provide educators with actionable insights to enhance instructional efficacy [5,6]. This technical review examines the architectural design, implementation strategies, and performance evaluation of AI-DSSs within online learning and assessment frameworks, highlighting their potential to revolutionize educational delivery and foster equitable, adaptive learning experiences.

1.1. Background and Motivation

The proliferation of online learning platforms, particularly massive open online courses (MOOCs), has been remarkable, with global enrollment growing at an annual rate of approximately 30% between 2020 and 2025 [7,8]. These platforms leverage LMSs to deliver course content, monitor learner progress, and administer assessments. However, traditional LMSs often rely on static, one-size-fits-all assessment methods, such as multiple-choice quizzes, which fail to adapt to individual learner profiles and struggle to scale for large cohorts [9]. Such limitations hinder the ability to provide tailored educational experiences and timely feedback, particularly in high-enrollment settings, as illustrated in Figure 1.

Traditional online learning and assessment systems face three primary challenges. First, scalability issues emerge when managing large learner cohorts, resulting in delays in feedback and grading processes [10]. Second, personalization remains limited, as uniform assessment criteria often overlook learners’ unique needs, learning paces, and preferences [11]. Third, manual grading of open-ended assessments, such as essays or project submissions, is labor-intensive and susceptible to subjective biases, compromising fairness and efficiency [12]. These challenges necessitate intelligent, automated solutions to enhance the adaptability and scalability of online learning environments.

AI-driven DSSs address these limitations by leveraging advanced computational techniques to process extensive learner data, including performance metrics, behavioral patterns, and demographic profiles. Machine learning algorithms, for instance, can predict learner outcomes and recommend personalized learning resources, while NLP techniques enable automated, high-accuracy scoring of open-ended assessments [13,14]. Knowledge-based systems further enhance decision-making by incorporating domain-specific rules to guide instructional strategies [15]. Collectively, these technologies improve assessment accuracy, enhance scalability, and promote equitable learning experiences, positioning AI-DSS as a cornerstone of modern educational technology [16], as compared in

Table 1. Comparison of traditional LMSs and AI-driven DSSs in online learning.

Feature	Traditional LMS	AI-Driven DSS
Scalability	Limited by manual processes	High, automated processing
Personalization	Static, uniform content	Adaptive, learner-specific
Feedback Speed	Delayed, manual grading	Real-time, automated
Assessment Type	Primarily multiple-choice	Diverse, including open-ended
Bias in Grading	Subjective, human-dependent	Objective, algorithm-driven

.

1.2. Objectives of the Review

This review provides a comprehensive, critical analysis of AI-driven DSS frameworks in the context of online learning and assessment systems. The specific objectives are as follows:

• To systematically evaluate the architectural components, AI techniques, and integration strategies of AI-DSS frameworks within online learning platforms, emphasizing scalability and adaptability.
• To assess design considerations, including data requirements, model selection, and user-centric features, alongside performance metrics such as accuracy, scalability, computational efficiency, and learner satisfaction.
• To identify technical, ethical, and practical challenges in AI-DSS implementation, including data privacy, algorithmic bias, and system interoperability, while proposing directions for future research to advance the field.

These objectives ensure a balanced exploration of theoretical foundations and practical implications, providing actionable insights for educators, system developers, and policymakers.

1.3. Scope and Methodology

This review focuses on AI techniques, including machine learning, system learning, natural language processing, and knowledge-based systems, applied to decision support systems in online learning and assessment frameworks. The scope includes platforms deployed between 2020 and 2025, encompassing MOOCs, institutional LMSs, and adaptive learning environments. Both formative (ongoing, feedback-driven) and summative (final, evaluative) assessment systems are examined, with an emphasis on their integration with AI-DSSs to optimize learning outcomes.

The methodology adheres to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines to ensure a rigorous and transparent review process, as summarized in

Table 2. Summary of systematic review methodology.

Component	Details
Databases	IEEE Xplore, Scopus, Web of Science
Keywords	“AI-based decision support,” “online learning,” “automated assessment,” “adaptive learning”
Search Period	January 2020–July 2025
Inclusion Criteria	Peer-reviewed journal articles, conference papers, technical reports; empirical evaluations or practical AI-DSS implementations
Exclusion Criteria	Non-peer-reviewed sources, studies lacking technical depth
Analysis Approach	Case studies, comparative analysis of AI-DSS implementations

. A systematic literature search was conducted across databases such as IEEE Xplore, Scopus, and Web of Science, using keywords including “AI-based decision support,” “online learning,” “automated assessment,” and “adaptive learning.” The search targeted peer-reviewed journal articles, conference papers, and technical reports published between January 2020 and July 2025. Inclusion criteria prioritized studies with empirical evaluations, novel AI-DSS frameworks, or practical implementations in real-world settings. Exclusion criteria eliminated non-peer-reviewed sources and studies lacking technical depth. Case studies of prominent platforms and comparative analyses of AI-DSS implementations provide practical context. Findings are synthesized to elucidate trends, challenges, and opportunities in the field.

2. Related Work

The integration of AI into online learning and assessment systems has significantly reshaped educational technology. This section synthesizes the evolution of online learning systems, examines the AI techniques underpinning decision support systems, and evaluates their practical applications to provide a robust foundation for understanding their design, impact, and limitations.

2.1. Evolution of Online Learning Systems

Online learning systems have come a long way since the early 2000s, picking up speed especially between 2020 and 2025. As shown in Figure 2, this timeline points out major steps in adopting AI-driven DSSs. Back then, tools like Blackboard and Moodle stuck to basic content sharing and hands-on grading, which worked fine for smaller groups but fell short when it came to handling bigger crowds or tailoring experiences to individual users [17]. Then came the COVID-19 outbreak, which really pushed the need for digital classes into overdrive and brought to light problems with expanding access, adjusting to different needs, and giving quick responses [18]. Fast forward to 2025, and sites like Coursera, edX, and Khan Academy started weaving in smart AI elements, using data insights to create flexible lessons and evaluations that boosted user involvement by as much as 25% [19,20].

The evolution of DSSs in education mirrors these advancements. Initially, DSSs were rule-based, supporting administrative tasks like course scheduling and enrollment management [21]. The integration of AI has expanded its capabilities to include predictive analytics, personalized feedback, and automated grading, driven by large-scale learner data [22]. This shift reflects a move toward data-driven, learner-centric education, where AI-DSSs utilize behavioral, performance, and demographic data to optimize outcomes.

Table 3. Evolution of online learning platforms and DSS features (2000–2025).

Period	Platform Features	DSS Features
2000–2010	Static content delivery, manual assessments	Rule-based systems for administrative tasks
2010–2020	Scalable MOOCs, basic personalization	Predictive analytics for enrollment and retention
2020–2025	AI-driven adaptive content, automated grading, real-time feedback	Advanced AI-DSS with predictive analytics, personalized feedback, and multimodal data integration

summarizes the evolution of platform and DSS features, highlighting the transition from rigid, standardized systems to dynamic, adaptive frameworks.

2.2. AI Techniques in Decision Support Systems

The effectiveness of AI-driven DSSs in online learning relies on a suite of computational techniques tailored to address specific educational challenges. ML is foundational, with supervised learning models, such as random forests and support vector machines, widely used to predict student performance based on historical data like quiz scores, engagement metrics, and course interactions [23,24]. Unsupervised learning, including K-means clustering, identifies learner archetypes to enable targeted interventions, while reinforcement learning optimizes dynamic learning paths by adapting content delivery in real-time based on learner feedback [25,26]. Reinforcement learning (RL), in particular, has shown promise in adaptive systems, where algorithms like Q-learning model learning as a Markov decision process to reward effective paths and penalize suboptimal ones, leading to 12–18% improvements in retention and performance in empirical studies [27,28]. However, critiques of existing RL applications reveal notable limitations. For instance, RL often suffers from sample inefficiency, requiring thousands of learner interactions to converge on optimal policies, which can be impractical in real-world educational settings with limited data [29]. Reward design poses another challenge; that is, poorly defined rewards may lead to unintended behaviors, such as overemphasizing short-term quiz scores at the expense of deep understanding, resulting in unstable training and up to 20-30% variability in outcomes across diverse cohorts [30]. These issues highlight the need for hybrid approaches, combining RL with supervised ML for more robust stability, as noted in recent implementations. Natural language processing (NLP) is critical for automating assessments and enhancing learner interactions. Transformer-based models, such as BERT, enable automated essay scoring with correlations to human grading exceeding 0.85, reducing subjectivity and grading time [31]. Emerging extensions of NLP include large language models (LLMs), like GPT variants, which generate personalized feedback or content, achieving up to 90% consistency in discussion analysis and curiosity measurement [32,33]. Yet, critiques of LLM applications in education underscore significant limitations. Hallucinations—fabricated or inaccurate outputs—occur in 15–25% of responses, undermining reliability in high-stakes assessments [34]. Additionally, inherent biases from training data can amplify inequities, such as undervaluing non-native language inputs, with fairness metrics showing disparate impacts of 0.20–0.30 in multicultural datasets. Computational demands further restrict deployment in resource-limited environments, often requiring specialized hardware that increases costs by 40–50% compared to traditional NLP. Addressing these through fine-tuning and bias audits is essential for broader adoption [35].

NLP-powered chatbots, deployed on platforms like edX, provide real-time student support, addressing queries with 90% accuracy in some implementations [36]. Knowledge-based systems, rooted in expert system paradigms, encode pedagogical rules to recommend tailored learning resources, achieving high precision in curriculum alignment [37,38].

Deep learning, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), has advanced adaptive assessment systems. These models analyze multimodal data—text, video interactions, and behavioral patterns—to assess learner proficiency with precision rates up to 92% [39]. Despite their accuracy, deep learning models are computationally intensive, posing challenges for resource-constrained environments. Collectively, these AI techniques enable DSSs to deliver scalable, accurate, and personalized educational experiences, as summarized in

Table 4. Comparison of AI Techniques in Online Learning Decision Support Systems.

Technique	Application	Strengths	Limitations
Machine learning	Performance prediction, content recommendation, reinforcement learning for adaptive paths	High accuracy (85–88%), scalable, dynamic adaptation (12–18% retention gains)	Requires large, clean datasets; RL sample inefficiency needs extensive interactions
Natural language processing	Automated essay scoring, chatbots, LLM-based feedback generation	Fast, reduces subjectivity (0.85 correlation), high consistency (90%)	Limited by linguistic complexity; LLMs risk hallucinations (15–25%), bias amplification
Knowledge-based systems	Personalized learning paths	Rule-based precision	Limited adaptability to new contexts
Deep learning	Adaptive assessments, multimodal analysis	High precision (92%), versatile	High computational cost

.

2.3. Existing AI-DSS Applications

AI-driven decision support systems have been widely adopted across online learning platforms, demonstrating significant improvements in personalization, scalability, and feedback delivery. This subsection evaluates key applications from 2020 to 2025, drawing on case studies, empirical research, and comparative analyses, as summarized in

Table 5. Part I—Summary of research papers on AI in online learning and assessment (2020–2025).

Paper	Method	Result	Model	Performance Quality
Chen et al. [40]	Supervised ML with regression analysis	Predicted student dropout with 88% accuracy	Random forest	High, validated across 500+ students
Ludwig et al. [31]	NLP with Transformer models	Achieved 0.85 correlation with human grading	BERT	High, reduced grading time by 70%
Amin et al. [26]	Reinforcement learning with Q-learning	Optimized learning paths, improved retention by 15%	Q-learning network	Moderate, scalable but resource-intensive
Sridharan and Akilashri [39]	Deep learning with CNNs	Assessed multimodal data with 92% precision	Convolutional neural network	High, but computationally demanding
Boulhrir and Hamash [41]	Systematic review of 71 studies	Identified AIEd trends in profiling and prediction	N/A	High, comprehensive synthesis
Kabudi et al. [42]	Content analysis of 434 papers	Highlighted adaptive systems’ effectiveness	N/A	High, broad coverage of AIEd
Šarić et al. [43]	Cluster analysis on behavioral data	Grouped learners for tailored interventions	K-Means Clustering	Moderate, effective for small cohorts
Lu and Cutumisu [44]	Neural network with backpropagation	Automated grading with 87% accuracy	Backpropagation Neural Network	High, efficient for large classes
Happer [45]	Mixed-method review	Emphasized feedback’s role in learning gains	N/A	High, balanced qualitative-quantitative insights
Yim and Wegerif [46]	Survey of 120 educators	Noted 25% engagement boost with AI tools	N/A	Moderate, subjective but insightful
Novais et al. [38]	Fuzzy expert system design	Assessed soft skills with 80% reliability	Fuzzy logic system	Moderate, reliable for specific skills
Bañeres et al. [47]	Predictive modeling with ML	Identified at-risk students with 85% accuracy	Logistic regression	High, practical for early intervention
Islam et al. [48]	Small-sample experiment with AI tool	Predicted performance with 83% accuracy	XGBoost	Moderate, limited by sample size
Hu et al. [49]	AI-enabled discussion analysis	Measured curiosity with 90% consistency	Discussion analytics model	High, reduced teacher workload
Morales-Chan et al. [50]	Generative AI for assignment design	Generated 100+ personalized tasks	GPT-based model	High, innovative but untested long-term

and

Table 6. Part II—Summary of research papers on AI in online learning and assessment (2020–2025).

Paper	Method	Result	Model	Performance Quality
Zhou et al. [51]	Deep learning with HCI integration	Assessed teaching quality at 7/10 score	Deep learning system	Moderate, needs refinement
Heil and Ifenthaler [52]	Systematic review of 138 studies	Found 18.8% used automated grading	N/A	High, robust evidence base
Shao et al. [53]	ML analytics on behavior data	Improved understanding of student patterns	Decision tree	High, actionable insights
Amrane-Cooper et al. [54]	Review of proctored exams	Highlighted AI’s role in 20% efficiency gain	N/A	Moderate, context-specific results
Akinwalere and Ivanov [55]	Learning analytics with predictive modeling	Identified at-risk learners with 89% accuracy	Gradient Boosting	High, effective for early intervention
Ma et al. [56]	Systematic review and bibliometric analysis	Mapped AI roles in language learning	N/A	High, extensive coverage of trends
Kuzminykh et al. [57]	Personalized feedback system design	Improved engagement by 30%	Adaptive feedback model	High, tailored to student demographics
Fodouop Kouam [58]	Experimental study on adaptive tutoring	Enhanced learning outcomes by 22%	Intelligent tutoring system	High, context-specific benefits
Susilawati [59]	Longitudinal bibliometric analysis	Traced AI evolution in engineering education	N/A	High, detailed historical insight
Yousuf and Wahid [60]	Multi-perspective analysis of AI tools	Highlighted learner, teacher, system benefits	N/A	High, broad conceptual framework
Demszky et al. [61]	Randomized controlled trial on feedback	Increased teacher uptake by 35%	NLP feedback model	High, scalable across courses
Liu and Chen [25]	Theoretical framework on AI assessment	Proposed real-time feedback systems	N/A	Moderate, theoretical but innovative
Ouyang et al. [62]	Empirical study on online HE AIEd	Improved retention by 18%	Adaptive learning system	High, validated in online settings
Popenici and Kerr [63]	Qualitative analysis of AI impact	Identified ethical and practical challenges	N/A	Moderate, insightful for policy

.

Coursera employs machine learning to recommend courses based on learner profiles, achieving up to 20% improvements in course completion rates [20]. edX leverages NLP for automated grading of open-ended responses, enabling scalable assessment for MOOCs with thousands of learners, with grading time reduced by 70% in some cases [31,64]. Institutional LMSs, such as those at Stanford and MIT, integrate knowledge-based systems with ML to deliver adaptive assessments aligned with specific curricula, enhancing learner satisfaction by 15% [65]. These applications align with the strengths and limitations outlined in Table 4, particularly ML’s scalability and NLP’s efficiency.

A comparative analysis, supported by 29 studies in Table 5 and Table 6, highlights key trends. Personalization, driven by ML and knowledge-based systems, improves retention by up to 20% [62,66]. For example, Chen et al. [40] used a random forest model to predict student dropout with 88% accuracy, validated across 500+ students. Scalability is evident in NLP systems, with Ludwig et al. [31] reporting a 0.85 correlation with human grading using BERT. Real-time feedback, enabled by deep learning, achieves 92% precision in multimodal assessments [39]. However, challenges include high computational costs and ethical concerns, such as data privacy and algorithmic bias [30,67].

Figure 2 illustrates the adoption timeline of AI-DSS, from rule-based systems in the 2000s to the post-2020 AI surge, driven by the pandemic and advancements in ML and NLP. Table 5 and Table 6 provide a comprehensive overview of recent research, covering methods like supervised ML, reinforcement learning, and systematic reviews, with performance metrics ranging from 80% to 92% accuracy. These findings underscore the maturity of AI-DSSs while identifying gaps in computational efficiency, bias mitigation, and ethical considerations, which the proposed framework in Section 3.1 aims to address.

3. Design Considerations for AI-Driven Decision Support Systems

The design of AI-driven DSSs for online learning and assessment requires a careful synthesis of technical robustness, pedagogical effectiveness, and user accessibility to address the demands of scalable, personalized education. As online learning platforms serve millions of learners globally, the architecture, data requirements, model selection, and user-centric features of AI-DSSs must be meticulously engineered to ensure scalability, accuracy, and inclusivity. This section explores these critical design considerations, synthesizing advancements from 2020 to 2025 to propose a comprehensive framework for developing effective AI-DSSs in educational contexts.

3.1. Proposed System Architecture

At the heart of any AI-powered DSS lies its architecture, which plays a crucial role in reshaping online education to make it more scalable, tailored to individuals, and fair for all users. Based on insights from studies published between 2020 and 2025, this review puts forward a flexible, modular AI-DSS design built around four main elements: gathering data, processing with AI, a decision-making core, and an intuitive user interface. As depicted in Figure 3, this setup blends smoothly into existing learning management systems (LMSs), enabling custom learning journeys, automatic evaluations, and instant feedback. In doing so, it tackles key shortcomings in expanding reach and customizing experiences, as highlighted in earlier research [5,68].

The data collection module aggregates diverse inputs to enable intelligent decision-making. These include student performance data (e.g., quiz scores, assignment grades), behavioral data (e.g., time on task, navigation patterns), demographic data (e.g., age, language proficiency), and course metadata (e.g., learning objectives, content outlines). These inputs, sourced directly from LMS platforms like Moodle or Canvas, provide a real-time snapshot of learner activity and context. To ensure data quality and compliance with regulations like GDPR and FERPA, the module incorporates preprocessing steps such as data cleaning, normalization, and anonymization [69,70]. Figure 3 depicts these inputs as distinct streams, enabling the system to handle heterogeneous data for large-scale platforms like MOOCs.

The AI processing module transforms raw inputs into actionable insights using a suite of techniques. ML algorithms, such as random forest and gradient boosting, predict learner outcomes with accuracies of 85–90% [71,72]. Natural language processing (NLP) models, including Transformer-based architectures like BERT, automate grading of written assignments, achieving correlations of 0.85–0.90 with human evaluators [31]. Knowledge-based systems apply rule-based logic to recommend curriculum adjustments, offering transparency critical for educational settings [38]. Deep learning, leveraging convolutional and recurrent neural networks, processes multimodal data (e.g., text, video interactions) for innovative assessments [39]. These techniques, shown as parallel processing streams in Figure 3, ensure robust and versatile analysis.

The decision engine synthesizes AI outputs to generate actionable recommendations, such as personalized learning paths, automated grades, and real-time feedback. Reinforcement learning optimizes these decisions by treating learner interactions as a dynamic sequence, improving outcomes by up to 12% [26]. To address concerns about bias and transparency, the engine incorporates explainability mechanisms [73]. The user interface delivers outputs through intuitive dashboards integrated into LMS platforms via standard APIs like Learning Tools Interoperability (LTI) [65]. Students access progress tracking and personalized recommendations, while instructors receive analytics and alerts for at-risk learners. A feedback loop, depicted in Figure 3, channels user interactions back to the data collection module, enabling continuous system improvement.

To underscore the distinctiveness,

Table 7. Comparison of the proposed AI-DSS architecture with existing systems.

System	Data Handling	AI Integration	Key Innovation	Superiority
Proposed AI-DSS	Federated learning	ML, NLP, RL hybrid	Dynamic feedback loop	Enhanced privacy (15% bias reduction); adaptive retention (18% gain)
Coursera	Centralized	ML recommendations	Profile-based personalization	Limited privacy; our model adds RL for dynamic paths
edX	LMS-integrated	NLP grading	Scalable assessments	Static; our hybrid improves multimodality and explainability

contrasts our architecture with established systems. Innovations include federated learning for decentralized, privacy-enhanced data handling (surpassing centralized models in Coursera by reducing breach risks) and a hybrid RL-ML decision engine for real-time path optimization (extending edX’s static NLP with 12–18% improved retention, as per empirical benchmarks [26,62]). This positions the framework as a novel synthesis advancing scalability and equity.

Implementation and Validation Roadmap

Bringing the proposed AI-driven decision support system (DSS) from concept to real-world application demands a carefully planned, step-by-step strategy that balances technical practicality with educational value. Insights from recent studies, such as those showing 12–18% performance improvements with adaptive reinforcement learning systems, inform a practical roadmap starting with prototype development. Leveraging open-source tools like TensorFlow for machine learning and natural language processing components, alongside frameworks like Flask or Django for seamless integration with learning management systems (via standards like Learning Tools Interoperability), the initial phase focuses on building a functional prototype. This prototype emphasizes essential tasks, such as preparing data to meet privacy requirements (e.g., anonymizing information to comply with GDPR) and training models using synthetic or publicly available datasets to mimic student interactions.

The next step involves testing the prototype in a controlled environment, such as a university course with 200–500 students, to assess its real-world performance. Evaluation during this phase combines measurable outcomes with user feedback. For instance, the system should achieve at least 85% accuracy in predicting student outcomes, validated through k-fold cross-validation, and maintain response times under 1.5 s, tested using tools like JMeter to simulate high user loads. Equally important are qualitative measures, such as student and instructor surveys targeting over 80% approval rates (using metrics like Net Promoter Scores) and feedback on usability from educators. To address challenges like reinforcement learning’s reliance on extensive data, hybrid approaches—combining human oversight with automated processes—can enhance reliability during early deployments.

Subsequent phases focus on refining the system through iterative testing, comparing the AI-DSS against traditional learning management systems in A/B trials to achieve measurable gains, such as 15–20% improvements in student retention. Broader testing across multiple institutions should incorporate diverse datasets, including those from Asian and European educational contexts, to minimize bias and ensure cultural relevance. Long-term evaluation over 6–12 months will track sustained benefits, such as enhanced academic outcomes. To overcome computational barriers, cloud platforms like AWS or Azure can provide scalable processing power, reducing reliance on local infrastructure. Throughout, ethical considerations remain central, with regular fairness audits (e.g., maintaining disparate impact ratios below 0.8) to ensure equitable and trustworthy application across varied online learning settings.

3.2. Data Requirements and Challenges

The efficacy of the proposed AI-DSS relies on diverse, high-quality data to drive predictive and personalized outputs.

Table 8. Data types and their roles in the proposed AI-driven DSS.

Data Type	Role in AI-DSS
Student performance data	Predict academic outcomes and identify at-risk learners for targeted interventions
Behavioral data	Analyze engagement and study habits to optimize content delivery and learning paths
Demographic data	Personalize interventions and ensure inclusivity for diverse learner populations

summarizes key data types—student performance, behavioral, and demographic—and their roles in the system. Performance data (e.g., quiz scores, exam results) enable predictive analytics, identifying at-risk learners with 85–90% accuracy [71,74]. Behavioral data (e.g., time on task, interaction frequency) inform engagement analysis and content adaptation [75,76]. Demographic data (e.g., age, language proficiency) ensure inclusivity by tailoring interventions to diverse learner profiles, such as multilingual feedback for non-native speakers [77,78].

Data-related challenges include privacy, quality, and scalability. Compliance with GDPR and FERPA requires robust anonymization and encryption protocols, integrated into the proposed model’s data pipeline [69,70]. Data quality issues, such as noise or bias, can undermine model performance, potentially leading to inequitable recommendations [73,79]. The proposed model mitigates this through preprocessing techniques like outlier detection and bias correction. Scalability challenges arise from the volume of data needed for robust AI models, particularly for resource-constrained institutions. Federated learning, processing data locally to preserve privacy, offers a viable solution, achieving comparable accuracy to centralized models [80,81]. The proposed AI-DSS adopts federated learning to balance scalability and privacy, aligning with recent advancements.

3.3. Model Selection Strategies

Selecting appropriate AI models is critical for effective AI-driven DSSs in online learning, balancing accuracy, interpretability, and computational efficiency.

Table 9. Comparison of AI models applicable for decision support systems in online learning.

Model	Application	Strengths	Limitations
Supervised machine learning	Performance prediction, student success forecasting	High prediction accuracy, scalable for various datasets	Requires large labeled datasets, potential bias in data
Natural language processing (Transformers)	Automated essay scoring, intelligent chatbots, feedback generation	High precision in language tasks, fast processing with contextual understanding	Struggles with complex semantics, requires large computational resources
Knowledge-based systems	Adaptive curriculum sequencing, rule-based tutoring systems	Transparent decision-making, interpretable rules	Limited adaptability to dynamic learning contexts
Deep learning models	Multimodal assessments, image/audio analysis, adaptive learning systems	Versatile across data types, high precision with large datasets	High computational cost, often lacks explainability

compares key AI models and their applications, strengths, and limitations. Based on literature from 2020 to 2025, this subsection outlines strategies for model selection in tasks like prediction, assessment, and recommendation. Supervised ML models, such as random forest and gradient boosting, excel in predictive tasks, achieving 85–90% accuracy in student performance forecasting [72,82]. NLP models, like BERT or RoBERTa, are optimal for text-based tasks, such as automated essay scoring, with correlations of 0.85–0.90 to human graders [31,83]. Knowledge-based systems provide interpretable recommendations for curriculum sequencing but lack adaptability to dynamic contexts [84]. Deep learning models, including convolutional and recurrent neural networks, support multimodal assessments but require significant computational resources [39].

3.4. User-Centric Design Principles

The proposed AI-driven decision support system (AI-DSS) prioritizes user-centric design to ensure accessibility, inclusivity, and usability for learners, instructors, and administrators. Figure 4 illustrates the user interface design, showcasing student and instructor dashboards. Drawing on literature from 2020 to 2025, this subsection outlines key design principles. For learners, accessibility features such as screen reader compatibility, adjustable text sizes, and multilingual interfaces address diverse needs, including visual impairments and non-native language proficiency [77,85]. To mitigate biases in AI recommendations, the system employs diverse training datasets and bias-aware algorithms [73].

Instructor dashboards provide real-time analytics, including at-risk learner alerts and performance trends, enabling timely interventions [65]. Administrator interfaces offer system-wide metrics, such as engagement and resource utilization, to support strategic decisions. The design emphasizes simplicity to prevent information overload, using co-design processes to align interfaces with stakeholder needs [85]. Integration with learning management systems (LMSs) via Learning Tools Interoperability (LTI) ensures seamless adoption [65]. By prioritizing user-centric design, the AI-DSS fosters equitable and engaging educational experiences, addressing gaps in accessibility and stakeholder engagement.

As shown in

Table 10. Summary of common evaluation metrics for AI-driven decision support systems in online learning.

Metric	Definition	Trade-Offs
Accuracy	Correct predictions ratio	High DL accuracy (92%) increases latency (30–60%); balance via hybrid models.
Response time	Feedback delay	Real-time (<1 s) trades accuracy in complex tasks; edge computing mitigates resource use.
User satisfaction	Survey-based utility	High personalization boosts (85%) but may reduce if bias persists.

, trade-offs are inherent in AI-DSS; for instance, deep learning’s precision often elevates computational demands, delaying responses in resource-limited settings. Conversely, simpler ML models ensure faster feedback but at lower accuracy. Real-time emphasis must account for consumption limits, addressed through optimization like cloud bursting for balanced performance.

4. Performance Metrics and Evaluation

The success of AI-driven decision support systems in online education and evaluation relies heavily on solid assessment approaches that gauge their technical capabilities, effects on users, and ability to scale. In recent years, the area has made notable strides in establishing and implementing key performance indicators, crafting thorough evaluation models, and comparing these AI tools to conventional methods. In this section, we bring together these progress points, delivering an in-depth examination of the indicators, assessment techniques, and comparative standards to inform the creation and rollout of AI-DSSs in teaching and learning environments.

4.1. Metrics for AI-DSSs

To gauge the effectiveness of AI-driven decision support systems (DSSs) in online education, a balanced evaluation approach is essential, blending technical performance measures with user-focused outcomes. For tools designed to predict student success or recommend tailored learning paths, key metrics include accuracy, precision, and recall. Accuracy measures the proportion of correct predictions, ensuring the system reliably forecasts outcomes like course completion. Precision evaluates how accurately the system identifies true positives, such as students likely to excel, while recall assesses its ability to capture all relevant cases, minimizing missed opportunities for intervention. These metrics are critical for fostering confidence in the system’s recommendations, particularly in high-stakes assessment contexts where inaccurate predictions could misguide support strategies.

Beyond predictive capabilities, metrics that reflect user experience and engagement provide deeper insights into the system’s impact on learners and educators. User satisfaction, often assessed through surveys exploring perceived usefulness and ease of use, draws on frameworks like the Technology Acceptance Model [86]. Engagement is measured through behaviors such as time spent interacting with AI-recommended resources or the completion rate of personalized tasks, offering evidence of the system’s ability to motivate and sustain learner involvement [87]. Equally vital are scalability and responsiveness, as AI-DSSs must handle large learner cohorts without performance degradation. Scalability assesses the system’s capacity to process growing data volumes while maintaining reliability, and response time measures the speed of delivering feedback, crucial for real-time applications [88].

Table 11. Summary of common evaluation metrics for AI-driven decision support systems in online learning.

Metric	Definition
Accuracy	The proportion of correct predictions made by the AI model over total predictions.
Precision	The ratio of true positive predictions to the total predicted positive cases.
Recall	The ratio of true positive predictions to the total actual positive cases.
User Satisfaction	The perceived usefulness, usability, and acceptance of the system, often measured through validated surveys or questionnaires.
Engagement	Behavioral indicators such as interaction frequency, task completion rates, or time-on-task, reflecting user involvement.
Scalability	The system’s ability to maintain effectiveness and efficiency when handling increased data volume or user load.
Response Time	The time taken by the system to generate feedback, results, or recommendations after user input.

summarizes these metrics and their definitions, providing a clear overview for evaluating AI-DSS performance.

4.2. Evaluation Frameworks

Evaluating the performance of AI-driven decision support systems in online education requires a thoughtful blend of numerical and experiential measures to capture their full impact. Quantitative approaches focus on robust performance metrics. For instance, k-fold cross-validation ensures models deliver consistent results across varied educational datasets, yielding reliable values for accuracy, precision, and recall [89]. Statistical methods, such as t-tests or ANOVA, allow direct comparisons between AI-DSSs and traditional systems, revealing gains in predictive accuracy or processing speed [90]. For predicting continuous outcomes, like student grades, error measures such as mean absolute error provide clear insights into the model’s precision, guiding refinements to enhance reliability in real-world educational settings.

Qualitative evaluations, on the other hand, focus on user perspectives through methods like surveys, interviews, or case studies. These approaches gather feedback from students and educators on aspects such as system usability, perceived equity, and educational impact [91]. Case studies, such as those examining AI-powered recommendation systems on platforms like Udemy or automated feedback tools in Canvas, provide contextualized insights into practical effectiveness. Such qualitative data are vital for identifying user-centered challenges, including interface usability issues or potential biases in AI outputs, which may not surface in quantitative analyses [92]. By combining these complementary approaches, evaluations achieve a balance of technical precision and practical relevance, ensuring AI-DSSs meet both performance and user needs.

4.3. Benchmarking

Benchmarking AI-DSSs against non-AI systems and across platforms provides a comparative perspective on their performance. Non-AI systems, such as traditional LMSs with manual grading, serve as baselines. Studies from 2020 to 2025 demonstrate that AI-DSSs outperform these systems in scalability and personalization. For example, NLP-based automated essay scoring achieves grading times up to 80% faster than human graders while maintaining comparable accuracy [93]. Similarly, ML-driven recommendation systems improve learner retention by 15–20% compared to static content delivery [76].

Cross-platform benchmarking reveals variations in AI-DSS performance, such as Coursera’s ML models excel in personalization but require significant computational resources, whereas edX’s NLP systems prioritize scalability for large-scale assessments. Custom institutional systems often lag in scalability due to resource constraints, but offer tailored solutions for specific curricula. Figure 5 illustrates a comparison of key metrics across these implementations, highlighting trade-offs between accuracy, response time, and scalability.

5. Challenges and Limitations

The deployment of AI-based decision support systems in online learning and assessment systems holds transformative potential, yet significant challenges and limitations accompany it. Between 2020 and 2025, researchers have identified technical, ethical, and implementation barriers that impede the widespread adoption and effectiveness of these systems. This section examines these challenges, offering a critical analysis of their implications and potential mitigation strategies to inform the design of robust AI-DSSs for educational contexts.

5.1. Technical Challenges

Deploying AI-driven decision support systems in educational settings presents significant technical hurdles, particularly in ensuring model interpretability and addressing embedded biases. Interpretability is vital in education, where instructors, students, and administrators need clear explanations for AI-driven decisions, such as predictions of academic performance or personalized learning recommendations [94]. Complex neural networks, despite their predictive strength, often lack the transparency needed to build trust, limiting their adoption in high-stakes educational contexts. Recent efforts have explored more interpretable alternatives, such as decision trees or rule-based systems, though these often sacrifice accuracy for clarity [95].

Equally critical is the challenge of algorithmic bias. When AI systems are trained on historical datasets, they risk perpetuating existing inequities, potentially favoring certain groups or marginalizing underrepresented students, which undermines fairness in education [96]. Additionally, scaling these systems to support large-scale digital learning platforms poses logistical challenges. Handling real-time data from thousands of users—such as student interactions or assessment results—demands substantial computational resources, often exceeding the infrastructure available at many institutions [97]. Addressing these barriers requires innovative approaches that balance predictive accuracy, interpretability, and scalability to unlock the full potential of AI-DSSs in transforming online education.

5.2. Ethical and Privacy Concerns

Incorporating AI-driven decision support systems into digital education environments introduces critical ethical and privacy dilemmas, especially in areas like data protection and the ramifications of automated judgments. Safeguarding data is essential, given that these platforms manage confidential details such as academic records and personal identifiers. Adhering to standards like Europe’s General Data Protection Regulation (GDPR) and the United States’ Family Educational Rights and Privacy Act (FERPA) is imperative, demanding strong measures for encryption and data de-identification [98]. Any lapses in security or improper handling could undermine confidence and subject educational organizations to substantial legal consequences.

Moreover, the moral dimensions of AI-facilitated decisions add layers of complexity to their implementation. These systems might unintentionally favor operational speed at the expense of justice, for example, by suggesting educational trajectories that marginalize particular student groups or employing skewed evaluation standards [96]. A case in point is natural language processing tools for grading essays, which, despite their speed, could overlook innovative or unconventional submissions if built on limited training data. To mitigate such issues, it is vital to develop ethical guidelines that emphasize fairness, openness, and responsibility, thereby guaranteeing that AI-DSSs promote educational equity rather than hinder it [83,99]. Bias detection employs metrics like equalized odds (threshold < 0.15 disparity) and correction via adversarial debiasing or dataset augmentation. Privacy guidelines: Collect only essential data with explicit consent; use differential privacy (epsilon < 1.0) for aggregation; store anonymized via tokenization [100]. Scenario: In a global MOOC with diverse learners, federated learning enables cross-cultural generalization, detecting language bias in NLP and correcting for 12% equity gain in non-Western cohorts. Real cases, like edX implementations, show that guidelines reduce breaches by 25%.

5.3. Implementation Barriers

Implementing AI-driven decision support systems (AI-DSSs) in online learning environments faces practical barriers related to cost, infrastructure, and adoption. Figure 6 illustrates the distribution of these challenges, categorized as technical, ethical, and implementation barriers. The development and maintenance of AI-DSSs require substantial financial investment, including costs for computational infrastructure, software licenses, and skilled personnel. Resource-constrained institutions, particularly in low-resource settings, often lack the budget to deploy advanced systems, limiting equitable access to AI-driven education [101,102,103]. Infrastructure requirements, such as high-performance servers and reliable internet connectivity, further exacerbate this challenge, especially in regions with limited technological resources.

Resistance to adoption by educators and institutions poses another key barrier. Educators may be skeptical of AI-DSSs due to concerns about job displacement or loss of pedagogical control [103]. Institutions may hesitate to invest in systems requiring extensive staff training.

Table 12. Mitigation strategies for challenges in AI-based decision support systems for online learning.

Challenge	Mitigation Strategy
Model Interpretability	Adopt explainable AI models (e.g., decision trees) and visualization tools
Algorithmic Bias	Use diverse, representative datasets and bias detection algorithms
Scalability	Implement cloud-based architectures and optimize computational efficiency
Data Security	Employ encryption, anonymization, and ensure compliance with GDPR/FERPA
Ethical Implications	Establish ethical design frameworks and include stakeholder input
Cost and Infrastructure	Utilize open source tools and cost-effective cloud infrastructure
Resistance to Adoption	Offer training and emphasize evidence-based pedagogical benefits

outlines strategies to address these challenges, including comprehensive training programs and stakeholder engagement to demonstrate the value of AI-DSSs in enhancing, rather than replacing, human expertise.

6. Case Studies

The practical application of AI-based decision support systems in online learning and assessment systems provides valuable insights into their efficacy, challenges, and potential for transforming education. Here we present two case studies, each illustrating distinct AI-DSS implementations in online learning environments. The first examines a Massive open online course (MOOC) platform employing natural language processing (NLP) for automated grading, while the second explores an adaptive learning system leveraging reinforcement learning for personalized learning paths. Through detailed analysis of their strengths, weaknesses, and outcomes, these case studies illuminate the real-world impact of AI-DSSs and inform future design considerations.

6.1. Case Study 1: AI-DSS in a MOOC Platform

Massive open online courses have democratized education, yet their scale demands efficient assessment solutions. A prominent MOOC platform, akin to edX, implemented an AI-DSS utilizing NLP for automated grading of open-ended responses, such as essays and short answers, in courses spanning 2021 to 2023. The system employed Transformer-based models, trained on extensive datasets of human-graded responses, to evaluate linguistic coherence, factual accuracy, and argument structure. Integrated within the platform’s learning management system (LMS), the AI-DSS provided immediate feedback to learners, a significant advancement over traditional manual grading processes [83,104,105].

The strengths of this implementation are manifold. The NLP system achieved grading accuracy comparable to human evaluators, with correlation coefficients exceeding 0.85 in validation studies, thereby reducing subjectivity and turnaround time [106]. Scalability was a key advantage, enabling the platform to process thousands of submissions simultaneously, supporting courses with enrollments exceeding 100,000 learners. Furthermore, immediate feedback enhanced learner engagement, as students could revise submissions based on AI-generated insights, fostering iterative learning [107].

However, weaknesses persisted. The system struggled with evaluating creative or non-standard responses, as its training data prioritized conventional answer structures, potentially stifling learner expression. Bias in training datasets also emerged as a concern, with the system occasionally undervaluing responses from non-native English speakers. Additionally, the computational cost of Transformer models posed challenges for real-time processing during peak usage periods [28,108]. Outcomes included a 70% reduction in grading time and a 15% increase in learner satisfaction, but addressing bias and computational efficiency remains critical for future iterations.

6.2. Case Study 2: Adaptive Learning System

Adaptive learning systems tailor educational experiences to individual learner needs, leveraging AI to optimize content delivery. A notable example is an institutional adaptive learning platform, deployed between 2022 and 2024, which utilized reinforcement learning to generate personalized learning paths [25]. The system modeled learner interactions as a Markov decision process, dynamically adjusting content difficulty and sequence based on performance and engagement metrics, such as quiz scores and time spent on tasks. Integrated with a custom LMS, the platform served undergraduate courses in STEM disciplines, targeting cohorts of approximately 5000 learners [104,109].

The reinforcement learning approach offered significant strengths. By continuously adapting to learner behavior, the system improved student performance, with average course grades increasing by 12% compared to static curricula [30]. Engagement metrics, including task completion rates, rose by 20%, reflecting heightened motivation among learners receiving tailored content. The system’s ability to identify at-risk learners early enabled timely interventions, reducing dropout rates by 10% [110]. Its integration with the LMS facilitated seamless instructor oversight, with dashboards providing actionable insights into learner progress. Nevertheless, the system faced challenges. The computational complexity of reinforcement learning required substantial infrastructure, limiting scalability for larger cohorts. Initial model training demanded extensive historical data, posing barriers for institutions with limited datasets. Moreover, some learners found the adaptive paths overly prescriptive, reducing perceived autonomy [111]. Outcomes demonstrated significant improvements in performance and engagement, but optimizing computational efficiency and learner agency is essential for broader adoption. Figure 7 illustrates the performance gains achieved by these AI-DSS implementations compared to traditional systems, and

Table 13. Summary of case study findings for AI-based decision support systems in online learning.

Platform	AI Technique	Outcome	Limitations
MOOC Platform	NLP (automated grading)	70% reduction in grading time; 15% increase in satisfaction	Bias in non-standard responses; high computational cost
Adaptive system	Reinforcement learning	12% grade improvement; 20% engagement increase	Computational complexity; limited learner autonomy

summarizes the findings from this case study.

7. Future Directions

The transformative potential of AI-based decision support systems in online learning and assessment is undeniable, yet their full realization depends on addressing current limitations and leveraging emerging opportunities. As the field evolves beyond 2025, advancements in AI technologies, strategies for scalability and accessibility, and the exploration of critical research gaps will shape the next generation of AI-DSS. This section outlines these future directions, drawing on trends from 2020 to 2025 to propose a roadmap for innovation and inquiry in educational technology.

7.1. Emerging AI Technologies

The rapid evolution of AI technologies offers promising avenues for enhancing AI-DSSs in online learning. Generative AI, critiques of existing LLM applications reveal limitations like output inconsistency (error rates 15–25% in factual tasks) and training biases amplifying cultural skews. RL in adaptive systems excels in personalization but suffers from exploration-exploitation imbalances, leading to suboptimal paths in sparse data scenarios (efficiency drops 20–30%). Our framework mitigates via hybrid integrations, blending RL with supervised ML for stability. These models can generate high-quality educational materials, including interactive tutorials, practice questions, and personalized feedback, tailored to individual learner needs. For instance, LLMs can produce contextually relevant explanations for complex topics, augmenting instructor efforts in large-scale courses. In assessment, generative AI can streamline automated essay scoring by evaluating not only linguistic accuracy but also conceptual depth, addressing limitations of earlier NLP systems. Multimodal AI, integrating text, images, and videos, represents another frontier. By analyzing diverse data inputs—such as learner-submitted diagrams, video-based responses, or interactive simulations—multimodal systems can provide richer assessments and recommendations. For example, a multimodal AI-DSS could evaluate a student’s video presentation for both content and delivery, offering holistic feedback. Recent advancements suggest that combining text and visual processing can improve assessment accuracy by up to 20% compared to text-only models. These technologies promise to enhance the adaptability and expressiveness of AI-DSS, fostering more engaging and comprehensive learning experiences.

7.2. Scalability and Accessibility

Expanding AI-driven decision support systems to a worldwide scale, especially in resource-constrained environments, represents both a significant hurdle and a promising avenue for progress. To achieve scalability, it is essential to enhance computational efficiency, allowing these systems to handle extensive user groups while maintaining high standards of performance. Utilizing cloud-based architectures, which incorporate distributed processing techniques, provides a practical solution by minimizing the need for on-site hardware. Additionally, freely available AI frameworks such as TensorFlow (version 2.16.1) and PyTorch (version 2.5.1) help reduce entry barriers, facilitating affordable creation and rollout for organizations operating under financial constraints.

Equally important is the focus on accessibility to promote fair access to learning opportunities across all demographics. AI-DSSs should be designed to support a wide array of users, encompassing individuals with disabilities, those speaking different languages, and participants from underprivileged areas. Progress in natural language processing enables the development of models that offer materials and responses in various languages, thereby overcoming communication obstacles. Inclusive elements, including compatibility with assistive technologies like screen readers and user-friendly designs, improve engagement for those facing visual or cognitive difficulties. Contemporary efforts stress collaborative design processes involving a broad range of participants to tailor systems to diverse requirements, particularly in areas with unreliable internet or limited device capabilities. Through these measures, the goal is to broaden the reach of AI-enhanced education on a global level.

7.3. Research Gaps

While AI-driven decision support systems have advanced considerably, several critical areas remain understudied, limiting their full potential in educational settings. One major gap lies in understanding the long-term effects on student success. Although short-term evaluations show gains in academic performance and engagement, extended studies are needed to examine enduring impacts on knowledge retention, skill acquisition, and professional achievements. Another significant shortfall is the lack of cross-cultural research. Current AI-DSS deployments largely reflect Western educational frameworks, potentially missing cultural factors that shape learning behaviors. For example, algorithms for personalized recommendations may require tailoring to accommodate preferences rooted in collectivist or individualist cultural perspectives across global regions.

Ethical dimensions of AI-DSSs also demand greater exploration. Key concerns, including bias reduction, equitable automated evaluations, and the psychological effects of AI feedback on learners, remain insufficiently addressed.

Table 14. Identified research gaps and recommended studies for AI-driven decision support systems in digital education.

Research Gap	Recommended Study
Long-term effects on student outcomes	Multi-year longitudinal research to monitor knowledge retention and career progression over 5–10 years
Cross-cultural relevance	Cross-regional comparative analyses to refine AI-DSS algorithms for diverse cultural settings
Ethical considerations	Studies on mitigating biases and assessing the psychological impacts of AI-generated feedback

outlines these gaps alongside proposed research to tackle them. Additionally, Figure 8 illustrates a forward-looking plan for AI-DSS advancements, highlighting key objectives from 2025 to 2031 across four domains: research (red), scalability (orange), technology (blue), and accessibility (green).

The technology domain begins with generative AI for creating customized educational content, such as interactive lessons and adaptive exercises, from 2025 to 2028. This progresses to multimodal AI systems between 2026 and 2030, integrating text, images, and videos to enable comprehensive evaluations, such as analyzing video-based student presentations for both content and delivery. By 2027–2031, the focus shifts to explainable AI, leveraging methods like SHAP or LIME to enhance transparency in decision-making processes, addressing current challenges in model interpretability. In terms of scalability, cloud-based solutions from 2025 to 2029 will enable efficient handling of large user groups through distributed computing, reducing reliance on local infrastructure and supporting real-time data processing for global platforms. From 2026 to 2031, open-source tools like TensorFlow and PyTorch will drive cost-effective development, encouraging collaborative innovation, particularly for institutions with limited resources.

Accessibility efforts prioritize inclusivity, with multilingual AI-DSSs from 2025 to 2028 using advanced natural language processing to provide content in multiple languages, addressing linguistic barriers for non-native speakers. Between 2026 and 2030, inclusive design will incorporate features like screen reader support and intuitive interfaces to accommodate learners with disabilities, enhancing access in underserved areas. Research priorities include longitudinal studies from 2025 to 2031 to evaluate long-term outcomes, filling the gap in understanding sustained impacts. Cross-cultural research from 2026 to 2030 will refine algorithms by analyzing implementations across diverse global contexts. Ethical frameworks, developed between 2027 and 2031, will emphasize bias audits, fairness metrics, and psychological assessments, advocating for regular equity reviews to minimize risks.

This structured roadmap aligns technological advancements with targeted research to address existing gaps, fostering equitable, scalable, and impactful AI-DSSs in digital education.

8. Conclusions

This comprehensive review of AI-based DSSs for online learning and assessment, spanning advancements from 2020 to 2025, underscores their transformative impact on educational technology. The integration of machine learning, natural language processing, knowledge-based systems, and deep learning has revolutionized online learning platforms by enabling personalized learning paths, automated assessments, and predictive analytics. Key findings highlight significant improvements in educational outcomes, with case studies demonstrating a 70% reduction in grading time for MOOC platforms and up to 20% increases in student engagement and performance through adaptive systems. The proposed system architecture—emphasizing data collection, AI processing, and seamless LMS integration—offers a robust framework for scalable and inclusive education delivery. As detailed in the roadmap in Figure 8, these steps provide a structured path for practical deployment and empirical validation. Performance metrics, including accuracy, response time, and user satisfaction, affirm the efficacy of AI-DSS, while evaluation frameworks combining quantitative and qualitative methods ensure rigorous assessment.

Despite these advancements, challenges persist, including model interpretability, algorithmic bias, data privacy concerns, and implementation barriers such as cost and adoption resistance. The transformative potential of the AI-DSS lies in its ability to democratize education, offering tailored learning experiences that cater to diverse global audiences. Looking forward, as outlined in Figure 8, emerging technologies like generative AI and multimodal systems, coupled with strategies for scalability and accessibility, promise to enhance educational equity further. Researchers and practitioners are encouraged to address critical gaps, particularly through longitudinal studies on learner outcomes, cross-cultural investigations, and the development of ethical frameworks to ensure fairness and transparency. By tackling these challenges and leveraging emerging opportunities, the academic and educational technology communities can advance AI-DSSs to create inclusive, efficient, and impactful online learning ecosystems.