Next Article in Journal
A New Method of Geological Modeling for the Hydrocarbon Secondary Migration Research
Previous Article in Journal
Municipal Sewage Sludge Disposal in the Republic of Poland
Previous Article in Special Issue
Development of a Short-Range Multispectral Camera Calibration Method for Geometric Image Correction and Health Assessment of Baby Crops in Greenhouses
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Applied Sciences
Volume 15
Issue 6
10.3390/app15063376
applsci-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

The Combination of Machine Learning Tools with the Rapid Visco Analyser (RVA) to Enhance the Analysis of Starchy Food Ingredients and Products

by
Joseph Robert Nastasi
1,
Shanmugam Alagappan
2 and
Daniel Cozzolino
2,*
1
School of Agriculture and Food Sustainability, The University of Queensland, St Lucia, QLD 4072, Australia
2
Queensland Alliance for Agriculture and Food Innovation (QAAFI), Centre for Nutrition and Food Sciences, The University of Queensland, St. Lucia Campus, Brisbane, QLD 4072, Australia
*
Author to whom correspondence should be addressed.
Appl. Sci. 2025, 15(6), 3376; https://doi.org/10.3390/app15063376
Submission received: 24 January 2025 / Revised: 15 March 2025 / Accepted: 17 March 2025 / Published: 19 March 2025
(This article belongs to the Special Issue Advances in Automation and Controls of Agri-Food Systems)
Browse Figure
Versions Notes

Abstract

:
This review discusses how the integration of machine learning (ML) tools enhances the analytical capabilities of the Rapid Visco Analyser (RVA), aiming to provide a deeper understanding of the starch gelatinization in different starchy food ingredients and products. The review also discusses some of the limitations of RVA as a tool for assessing the pasting and viscosity behavior of starch, emphasizing the potential of different ML tools such as principal component analysis (PCA) and partial least squares (PLS) regression to offer a better analytical approach. Examples of the utilization of ML combined with RVA to enhance the analysis of starch and non-starch ingredients are also provided. Furthermore, the importance of preprocessing techniques, such as derivatives, to improve the quality and interpretability of RVA profiles is discussed. The aim of this review is to provide examples of the utilization of RVA combined with ML tools in starchy food ingredients and products.

1. Introduction

Starch is found in a wide range of seeds (e.g., cereal grains, pulses) and tubers from different agricultural crops (e.g., maize, wheat, rice, barley, sweet potato, and cassava), and it serves as the main source of energy in the human diet [1]. Moreover, this polysaccharide is also a key component in the formulation of animal diets (mainly chicken and pigs), as it provides a source of highly digestible energy [2]. Starch plays a significant role in the texture, stability, functional properties, and digestibility of starchy food ingredients and food products [3,4]. As a result, its influence extends beyond nutrition, contributing to food and feed functionality, making it a critical nutrient in food and nutrition [3,4].
The biochemical characteristics and rheological properties of starch are associated with the proportion, chain length, and structure of both amylose (AMY) and amylopectin (AMP) [3,5,6,7,8]. Both AMY and AMP influence the viscosity or gelatinization properties of the starch, changing the functional properties of the food ingredient or product, as well as its utilization (e.g., bread making, malt and beer properties, sauces, or in biopolymer formulations) [3,5,6,7,8].
The retrogradation of starch is a process that is associated with the disaggregation of the AMY and AMP chains that are present in the gelatinized starch paste or sample, as well as their reassociation to form a more ordered structure [8,9,10,11]. Other starch properties that influence gelatinization and retrogradation are associated with the characteristics of the starch granule (e.g., type of granule, particle size) such as swelling, and breakdown, contributing to the texture and stability of starchy food ingredients and products [12].
The Rapid Visco Analyser (RVA) is a widely used instrument for measuring the rheological properties of starch and starchy ingredients under controlled heating and cooling conditions. By providing viscosity profiles, the RVA helps researchers and food scientists evaluate starch’s behavior in food and industrial applications [10]. However, while the RVA is efficient in capturing pasting properties, its interpretation is often limited to conventional parameters such as the pasting temperature (PT), peak viscosity (PV), breakdown (BD), setback (SB), and final viscosity (FV) [10]. These standard parameters, although useful, do not fully explain the complexity of starch’s structure–function relationships. Additionally, variability in starch properties due to differences in botanical origin, processing methods, and environmental factors can make RVA data interpretation challenging [8].
A key limitation of RVA-based starch analysis is its reliance on traditional statistical approaches that often fail to capture hidden patterns, nonlinear relationships, and multivariate interactions in viscosity profiles. Conventional analysis typically considers only a few isolated parameters, potentially overlooking critical insights into starch’s gelatinization and retrogradation mechanisms. Furthermore, inconsistencies in RVA testing protocols across different laboratories can lead to variability in reported results, impacting the reproducibility and comparability of the data [9,10].
To overcome these limitations, the integration of machine learning (ML) techniques, such as principal component analysis (PCA) and partial least squares (PLS) regression, provides a more comprehensive approach to RVA data analysis [10]. These ML tools enable the identification of dominant trends, classification of starch types based on their pasting behavior, and predictive modeling of starch’s functionality. PCA is particularly useful for reducing data dimensionality and visualizing trends, while PLS enhances the predictive capabilities of RVA by correlating pasting properties with compositional attributes. By leveraging ML, RVA analysis can transition from a purely empirical approach to a more data-driven methodology, offering deeper insights into starch’s behavior in food formulations.
The significance of this research extends beyond academic inquiry, as improved RVA data interpretation can lead to advancements in food quality control, ingredient selection, and starch-based product development. The ability to classify starch sources, optimize formulations, and predict functional outcomes has direct implications for the food manufacturing, bakery, brewing, and biopolymer industries. This review aims to explore how ML-enhanced RVA analysis can bridge existing knowledge gaps in the characterization of starch, ultimately contributing to more efficient, reliable, and standardized starch evaluation methodologies in both research and industry applications.

2. Rheology of Starch—Its Measurement, Instrumentation, and Limitations

As stated in the previous section, starch can be defined as a complex biopolymer material due to its ability to undergo physical changes, such as swelling, gelatinization, and retrogradation [13,14]. The study of how materials like starch flow and deform under various physical forces, such as shear or compression, is known as rheology [13,14].
The measurement of the rheological characteristics or properties of a sample implies that a controlled deformation or strain is applied to a material over a given time [14]. The resulting force response from this process is measured. This response provides an indication of different parameters and characteristics of the material or sample such as the stiffness, modulus, viscosity, hardness, strength, or toughness of the material [14]. Consequently, rheology is utilized in different ways to quantify and describe the mechanical properties of a given material, to collect information associated with the molecular structure and composition of such material, to characterize and simulate the performance of a given material during processing, or to predict the final product quality of the sample [8,15,16,17,18].
Techniques that are used to measure rheological properties in cereals have been classified into descriptive empirical techniques or those based on fundamental measurements [8,15,16,17,18]. Techniques classified as fundamental rheological tests are based on the grounds that the sample geometry is constant and well defined, with both the stress and strain states of the samples being controlled and uniform [15,16,17,18]. These measurements require sophisticated and expensive instrumentation that is time-consuming, difficult to maintain in an industrial environment, and requires high levels of technical skills [15,16,17,18].
In starch rheology, the measurement of a starch’s pasting properties (e.g., cereals, starchy foods, and other polysaccharides) is the result of cooking starch in an excess of water [19]. This principle is utilized by most commercial viscometers, including the RVA. This instrument operates by applying a controlled shear force to the sample while simultaneously subjecting it to precise thermal cycling [19]. The RVA is not the only instrument available, where other practical rheological methods (e.g., Brabender visco-amylograph and Brookfield Viscometer) have been extensively utilized in both cereal and starch research to evaluate both the gelatinization and the pasting properties of the samples associated with the functional properties of the starch [19,20,21]. The RVA is preferred over the Brabender visco-amylograph for its faster heating and cooling rates, smaller sample size, and higher sensitivity to viscosity changes, ensuring more precise and reproducible results for starch pasting and gelation analysis. Unlike the Brookfield Viscometer, which measures viscosity at a constant shear rate and is suited for steady-state assessments, the RVA captures dynamic pasting properties such as gelatinization, peak viscosity, breakdown, and setback. This comprehensive analysis makes the RVA particularly valuable for food formulation, quality control, and ingredient performance testing.
However, the application of these empirical methods is often associated with some drawbacks that can be considered limiting factors in some food applications [19,20,21]. These drawbacks can be associated with a poor or inconsistent definition of both the measured parameters or protocols that are utilized in each laboratory, the time-consuming nature of some of the current methods (e.g., the time required for sample preparation), the difficulties in the interpretation of the results (e.g., results based on the interpretation of a single parameter or the effect of moisture), as well as issues associated with sample requirements (e.g., the number of replicate samples, sample particle size—including preprocessing of the sample prior to analysis, such as grinding—or the effect of the temperature or harvest conditions) [15,16,17,18,19,20,21].
Figure 1 illustrates the typical RVA profile obtained after analyzing different rice flour samples with varying concentrations of AMY. It has been observed that the start and end of each stage or step in the RVA profile depends on the temperature gradient (black dotted line) of the pasting method utilized. Overall, a standard RVA profile yields the following parameters: the pasting temperature (PT), peak viscosity (PV), Time to Peak (PTi), breakdown (BD), Trough/Minimum Viscosity (TV), setback (STB), and final viscosity (FV) (Figure 1) [15,16,17,18]. While the definitions and significance of these parameters remain consistent among researchers and applications of this technique, different abbreviations are often used, and some of these are summarized in Table 1.

3. Machine Learning Tools

Enhancing the information generated by the RVA beyond the interpretation of a single parameter (e.g., PV or FV) requires the utilization of data analytics tools, including techniques and methods that are available in the fields of chemometrics and ML [22,23,24]. The utilization of ML tools to analyze an RVA profile has allowed for the determination of the geographical origin of a sample (e.g., region, country), to identify the different varieties or species (e.g., cereals), or to monitor the process that a given sample had undergone (e.g., heating). Specifically, the utilization of methods such as PCA or cluster analysis (CA) applied to the RVA profile has been reported by different authors [16,18].
PCA is an unsupervised machine learning technique used to reduce the dimensionality of data while preserving as much variance as possible. It transforms a large set of correlated variables into a smaller set of uncorrelated variables called principal components (PCs), which retain the most significant patterns in the dataset. In RVA analysis, PCA is widely used to classify starch sources based on their pasting properties, detect variability among starch samples, and visualize trends in large datasets [16,18]. The primary advantage of PCA is its ability to highlight dominant trends in RVA profiles without requiring prior knowledge about sample classification. However, it does not establish causal relationships between pasting properties and starch composition, making it more suitable for exploratory data analysis rather than predictive modeling. Additionally, PCA requires proper data preprocessing (e.g., centering and scaling) to ensure meaningful component extraction, and the interpretation of PCs may sometimes be challenging when multiple factors influence the RVA viscosity profiles simultaneously.
Other machine learning tools have been used to develop quantitative models of the relationship between the RVA profile and other chemical properties, where the incorporation of algorithms such as PLS regression are the most utilized [16,17,18]. PLS is a supervised machine learning technique designed to establish a predictive relationship between input variables (e.g., RVA pasting parameters) and response variables (e.g., amylose content, starch composition). Unlike PCA, which is primarily exploratory, PLS identifies latent variables that maximize the covariance between independent and dependent variables, making it particularly useful for predicting functional and compositional attributes from RVA profiles [16,18]. The advantage of PLS is its ability to handle collinear data, extract meaningful relationships between RVA parameters and starch properties, and improve model interpretability. However, PLS models require proper validation (e.g., cross-validation or independent validation) to prevent overfitting, especially when dealing with small sample sizes. Additionally, while PLS improves the predictive accuracy, it still relies on well-defined input–output relationships and can be sensitive to noise in the dataset.
In recent years, other algorithms have also been explored and reported to analyze the RVA profile such as support vector machines (SVMs), either as a classification or regression tool, artificial neural networks (ANNs), K-Nearest neighbors (KNNs), or convolutional neural networks (CNNs) [16,17,18,25,26,27,28,29,30,31,32,33,34,35,36,37,38]. In the context of the RVA, ML techniques such as PCA, PLS, and PLS discriminant analysis (DA) have been used the most routinely, as demonstrated in Table 2, and appear to be the most relevant for research applications.
Please note that it is beyond the objective of this review to provide detailed information about these algorithms, on which detailed information can be found elsewhere [22,23,24]. It is important to highlight that during the application or implementation of ML tools, not only is the selection of the algorithm important (e.g., PLS or ANN), but also, other variables such as the sampling method, the selection of samples and replicates, as well as the validation method used (e.g., cross-validation vs. independent validation) are key elements that need to be considered during the development and interpretation of the ML models.

4. Examples of Combining ML Tools to Interpret the RVA Profile

A summary of the studies that have reported the development of applications by combining ML tools with RVA is provided in Table 2. Some of these applications will be discussed in more detail in the following sections.
Wongsaipun et al. (2018) [25] reported the utilization of RVA to evaluate waxy and non-waxy rice grain samples stored in paddy form at an ambient room temperature (28–32 °C) for 1 year. In this study, the RVA profiles of the rice samples were recorded every month, and the data were analyzed using both partial least squares (PLS) regression and supervised self-organizing maps (supervised SOMs) to predict the time of storage of the sample. The PLS regression models developed using the RVA profiles resulted in a root mean square error of cross-validation (RMSECV) of 1.2 and a coefficient of determination in cross-validation (R2) of 0.90, with a ratio of prediction to deviation (RPD) of 3.2. The PLS variable influence on projection (PLS-VIP) indicated that PV and FV were the parameters with a strong influence on the PLS models that were used to predict the time of storage [25]. The achieved R² value of 0.90 indicates a strong correlation between the predicted and actual values, suggesting that the model effectively captures the underlying patterns in RVA pasting properties in relation to storage. Given the inherent variability in starch gelatinization and pasting behavior due to differences in botanical origin, processing conditions, and moisture content, an R² of 0.90 is considered highly acceptable for food-based regression models. Additionally, the RMSECV value of 1.2 demonstrates the model’s robustness in minimizing prediction errors across different storage conditions. While direct comparisons with other studies may vary due to differences in experimental design, sample composition, and preprocessing techniques, the performance achieved here aligns with the general expectations for PLS models when applied in food science. It is important to note that variations in the data distribution, feature selection, and noise levels can all influence the model accuracy. In particular, the use of derivative preprocessing and outlier removal likely contributed to enhanced model reliability, which is discussed in the next section of this review.
To assure the safety and origin of the ingredients used, the malting, beer, and whisky industries are required to implement traceability systems [16,17,26]. Considering that starch is the main component in barley grain, any change in its biophysical or biochemical properties will influence the functional and malting properties of the sample (e.g., endosperm breakdown, protein content, cell walls, AMY and AMP contents and ratios, starch granule structure, lipids, and particle size) [16,17,26]. The ability of RVA combined with principal component analysis (PCA) and PLS-DA was evaluated to identify the origin of commercial barley varieties gown in different regions of South Australia. The PLS-DA models correctly classified 96.3 and 97.8% of the samples according to harvest and locality. Furthermore, the interpretation of both the PCA and PLS loadings provided critical information that was hidden in the RVA profile that was used to better explain which starch pasting properties of the sample contributed to explaining the differences between the different regions [16,17,26].
A study by Kaur et al. (2013) [27] evaluated the physicochemical properties and RVA profiles of taro (Colocasia esculenta L.) using Pearson correlation and PCA. The results from this analysis demonstrated differences between taro flour and other flour samples from different crops. The taro samples had high carbohydrate contents and water absorption levels and lower protein contents, foaming capacities, and STBs [27]. The PV of taro flour was lower in comparison to potato flour and higher than soya and corn flours. The visualization of the scores derived from the PCA indicated that taro and potato flours had negative scores, whereas the soybean and corn flour samples had positive scores along the first principal component [27].
Recently, a study by Zhu et al. (2018) [28] demonstrated the ability of RVA profiles, combined with ML tools, to identify viscosity fingerprints that can serve as profile markers for the classification of rice from different regions and varieties. The full RVA profiles obtained from 152 rice samples were analyzed using PCA and PLS-DA [28]. The results showed that different rice subspecies or group types could be distinguished by interpreting the RVA profile in combination with PCA.
The ability of RVA profiles combined with ML tools to obtain useful information for the traceability of ingredients in alternative protein-based formulations has been reported [35]. RVA was used to evaluate the addition of cricket powder to chickpea and flaxseed flours via PCA and PLS regression [35]. The utilization of PCA and PLS regression enabled both qualitative and quantitative identification of the addition level of cricket powder to chickpea and flaxseed flour samples. Differences in the PLS loadings associated with the RVA profile due to the addition of CKP was reported by the authors, and the cross-validation statistics for the predicted level of cricket flour were as follows: R2CV = 0.90–0.95 and SECV = 5.7–7.7. This study demonstrated that integrating RVA profiles with machine learning tools provides a reliable, accurate method for ensuring proper ingredient formulation, which is crucial for both quality control and meeting the traceability demands in alternative protein-based food products [35].

5. Preprocessing of RVA Profiles—Effect of Derivatives

The utilization of ML tools requires some level of processing of the RVA profile before analysis. Several authors showed how the utilization of derivatives or smoothing of the RVA profile enhances the quality of the profile [16,17,37,38]. For example, to better understand the pasting properties of rice, the RVA of the grain was collected, and the profile was interpreted using derivatives [37]. The use of the first derivative as preprocessing of the RVA profiles assisted the researchers in visualizing differences in the pasting properties of the sample that were not easily detected by interpreting the raw RVA profile [37]. These authors reported that samples containing low AMY contents (varieties with 0.41% amylose and 7.04% protein) exhibited a single smooth transition in the RVA profile during pasting. Furthermore, the pasting properties of all the other samples analyzed [e.g., M201 (TX), Nato (LA), Koshihikari (CA), Mercury (LA), and Nanking Sel (LA)] with higher amylose contents (10.65–24.9%) underwent multiple phase transitions and rate changes before the PV, as observed in the RVA profile after the use of the first derivative [37].
The relationship between AMY and AMP, as well as their influence on the functional properties of both maize starch blends and “model mixtures” made with waxy and high-AMY maize starch, was also reported [38]. In this study, the first derivative of the RVA profiles assisted the authors in identifying four parameters, namely Peak 2, Peak Time 2, Peak 3, and Peak Time 3 [38]. The authors reported that the PV was negatively correlated with these parameters, as well as with the AMY content, allowing them to conclude that AMP can modulate the water uptake of the sample [38]. The high FV values decreased as either AMP or AMY was added to the mixtures, suggesting that FV is the most sensitive parameter to interactions in starch [38]. The first derivative of the RVA profiles, as well as the alternative parameters, seemed to be suitable for demonstrating the importance of these interactions between the starch components with other parameters during gelatinization [38]. Preprocessing the RVA profile (using derivatives) before the use of PCA and PLS regression to evaluate the starch pasting characteristics of selected barley materials from a breeding program has also been reported [16]. The use of both PCA and PLS regression combined with the preprocessing of the RVA profile using derivatives (e.g., first and second derivatives) improved the interpretation of the profile, resulting in more analytical information relating to the pasting properties of the sample [16].

6. Advantages and Limitations of ML Combined with RVA

The RVA has been widely used in both research and the food industry, where the interpretation of RVA data is generally straightforward and simple by calculating single parameters (e.g., PV, FV) [10]. However, the information provided by the RVA instrument can also be considered as generic and limited, where several factors, such as the particle size and moisture content, can affect the calculated RVA parameters [8,9]. Some of these issues are known to contribute to the variability and reproducibility of the RVA instrument compared to other techniques [8,9,10]. Consequently, the definition of protocols and methods that incorporate these issues are of importance to compare the results from different laboratories or applications [8,9,10].
The application of ML tools to instrumental techniques such as the RVA is considered by some researchers as a mere exercise conducted by data scientists or modeling experts, where the development of the so-called complex algorithms and their application to simple instruments such as the RVA do not contribute with a better understanding of the pasting properties of the starch and starchy foods. However, the analysis of the RVA profile by incorporating ML tools (e.g., PCA, PLS, ANN) into the interpretation of the results has provided us with the means to classify samples according to species, variety, and processing or even to detect contamination [31,35,39,41].
PCA has been particularly valuable in RVA data analysis by identifying dominant patterns and clustering samples with similar pasting properties. Its advantages lie in its ability to simplify high-dimensional data, provide insights into sample variability, and serve as an effective exploratory tool for detecting quality variations in starch. However, a key limitation of PCA is that it does not directly model relationships between RVA parameters and specific starch properties. It only reduces the data complexity, making it less suitable for tasks requiring predictive modeling or a detailed mechanistic understanding of starch’s behavior. PLS, on the other hand, is a powerful tool for predictive modeling using RVA data. It allows researchers to establish robust relationships between pasting properties and the compositional or functional characteristics of starch-based ingredients. PLS is particularly useful in food formulation, enabling ingredient optimization based on functional properties derived from RVA analysis. Furthermore, PLS models are highly dependent on the dataset quality and require careful validation to prevent overfitting. Unlike PCA, which provides easy-to-interpret visual patterns, PLS involves a more complex interpretation of latent variables and requires appropriate sample selection to ensure model generalizability.
Beyond the utilization of a given algorithm to analyze the data, the preprocessing of the RVA profile using derivatives has provided a new dimension for the interpretation of the RVA profile [37,38]. The first derivative contributes to eliminating a constant offset, while the second derivative removes both the offset and the linear term (e.g., linear fitting applied to the spectrum) [37,38]. In some cases, the second-order derivative can accentuate sharp features in the profile and contributes to resolving overlapping profiles [37,38]. However, limitations in the use of derivatives are associated with the fact that high-order derivatives (e.g., second and third derivatives) can be more sensitive to instrumental noise or unwanted artifacts during the analysis compared with lower-order ones (first derivatives) [37,38,42].
Overall, the use of derivatives has provided researchers and analysts with a tool to better understand the pasting properties of starch, as well as to better interpret complex interactions between starch and other components during the analysis beyond the interpretation of a single parameter that is provided by the RVA instrument (e.g., FV). However, further research within this area is necessary to assist in maintaining the relevance of the RVA as an analytical instrument for both research and industry applications.
While PCA is valuable for extracting dominant patterns in RVA profiles, it does not establish causal relationships between pasting properties and starch’s composition. PLS overcomes this limitation by constructing predictive models, but overfitting remains a concern when models rely on insufficient training data. Additionally, PCA and PLS require careful data preprocessing, such as centering and scaling, to ensure meaningful results. Understanding these constraints is critical for applying ML methods effectively to RVA analysis.

7. Commercial Considerations Regarding ML in the Food Industry

The integration of ML tools into RVA analysis has significant implications for commercial applications, particularly in the areas of quality control (QC) and online automation in food manufacturing. Many industries, such as the snack industry, rely on rapid and accurate starch characterization for ensuring product consistency and functional performance [43]. ML-powered data analysis can enhance decision-making by providing real-time insights into ingredient functionality and formulation optimization.
Advancements in cloud computing and Internet of Things technologies have facilitated the adoption of ML-driven platforms in food production environments [44]. Several cloud-based platforms allow for real-time monitoring of RVA profiles, automatic anomaly detection, and predictive quality assurance [45]. Examples include AI-driven software that integrates with laboratory instruments to streamline data acquisition and processing, providing manufacturers with automated alerts when deviations from the expected starch behavior occur.
The implementation of ML-enhanced RVA analysis in commercial food production will drive greater efficiency, reduce operational risks, and improve product consistency. Future developments may include the application of deep learning models and automated machine learning frameworks to further refine predictive analytics for food texture and ingredient stability [46]. These advancements highlight the transformative potential of ML in modernizing starch quality assessments in both research and industry applications. Recently, Perten (Perking-Elmer) has included in the Prediction Pack an optional feature that is accessible to all users of the Thermocline for Windows (TCW) software for the RVA, the doughLAB for Windows (DLW) software for the doughLAB and micro-doughLAB, and the TexCalc software for the TVT [47]. This feature allows The Prediction Pack to import PCA or PLS models created in The Unscrambler software to predict a desired parameter [47].
Despite its potential, the integration of ML with RVA analysis presents several challenges that need to be addressed for successful implementation in both research and industrial settings. One of the key challenges is the availability of large, high-quality datasets for ML training and validation, as many food industry datasets remain proprietary, limiting broader applications. Additionally, the computational requirements for advanced ML models, such as deep learning, can be a barrier for small to medium-sized enterprises that lack infrastructure for large-scale data processing. While cloud-based ML solutions provide an alternative, concerns regarding cost and data security persist. Another critical issue is the lack of standardized ML methods for RVA data interpretation. Variability in preprocessing techniques, feature selection, and model validation methods can result in inconsistent results across different studies and applications. Addressing these challenges requires the development of standardized ML pipelines, greater access to diverse datasets, and optimization of computational efficiency to enable real-time quality control in food manufacturing.

8. Final Considerations

Different researchers have shown that starch’s pasting properties can be measured using empirical methods such as the RVA. The analysis of RVA profiles combined with ML tools has shown that is possible to use not only a single parameter (e.g., PV or FV) but also to interpret changes in the profile associated with the gelatinization of the starch, as well as their effect on the properties (e.g., functional properties, interactions between chemical components including lipids and water) of the analyzed food ingredient and food product.
Most of the applications that are discussed in the scientific literature on the analysis of cereals and starchy foods using RVA have reported and interpreted single parameters derived from the RVA profile (e.g., PV, PT, and FV). Although this approach is considered the routine application of the RVA to gather information about starch’s gelatinization or the pasting properties of a sample, other information about the starch’s pasting characteristics is lost during the analysis, weakening the ability of the RVA to be a valuable analytical tool. Consequently, the utilization and incorporation of ML tools (e.g., algorithms and preprocessing) has been shown to enhance the information that is generated by this type of instrument.
Therefore, the routine application of the RVA instrument to evaluate the gelatinization and retrogradation of starch can be enhanced by the incorporation of ML tools. This approach will enable a greater understanding of the rheological or pasting properties of starch by analyzing and interpreting the whole RVA viscosity profile. Finally, the utilization of pasting properties and empirical methods such as the RVA need further development, where research and development will play a key role to better understand the gelatinization and retrogradation of starch and starchy foods.
The advantage of machine learning (ML) is not solely focused on developing a model to predict a single parameter (e.g., PV). Rather, the overall benefits of these tools lie in their ability to mine data from the RVA trace and use it as a fingerprint. Furthermore, the RVA fingerprint, when combined with ML tools, can identify trends or patterns in the dataset or group of samples. This can be used to classify or identify samples based on AMY or AMP content, distinguish between waxy and non-waxy rice samples, predict storage duration in starchy foods, determine the geographical origin of barley and rice samples, identify the type of starch in different flour types, and detect the addition of insects to cereal flour, among other applications. Without the integration of the RVA profile with multivariate analysis (MVA) and ML tools, these applications would not be possible through the sole interpretation of a single RVA value.

Author Contributions

Conceptualization, D.C. and J.R.N.; methodology, D.C.; software, S.A., D.C. and J.R.N.; validation, S.A., D.C. and J.R.N.; formal analysis, D.C. and J.R.N.; investigation, J.R.N. and D.C.; resources, S.A., D.C. and J.R.N.; data curation, D.C. and J.R.N.; writing—original draft preparation, D.C. and J.R.N.; writing—review and editing, S.A., D.C. and J.R.N.; visualization, J.R.N. and D.C.; supervision, D.C.; project administration, D.C. All authors have read and agreed to the published version of the manuscript.

Funding

The support of the University of Queensland and Queensland Alliance for Agriculture and Food Innovation is acknowledged. Joseph Robert Nastasi acknowledges the support of the Australian Research Council: ‘A Deadly Solution: Towards an Indigenous-led bush food industry’, grant ID: GA141113.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The datasets used and analyzed within the study may be obtained from the corresponding author upon reasonable request.

Conflicts of Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References

  1. Hardy, K.; Brand-Miller, J.; Brown, K.D.; Thomas, M.G.; Copeland, L. The importance of dietary carbohydrate in human evolution. Q. Rev. Biol. 2015, 90, 251–268. [Google Scholar] [CrossRef] [PubMed]
  2. Svihus, B.; Uhlen, A.K.; Harstad, O.M. Effect of starch granule structure, associated components and processing on nutritive value of cereal starch: A review. Anim. Feed Sci. Technol. 2005, 122, 303–320. [Google Scholar] [CrossRef]
  3. Schwartz, D.; Whistler, R.L. History and future of starch. In Starch: Chemistry and Technology, 3rd ed.; BeMiller, J.N., Whistler, R.L., Eds.; Academic Press: Cambridge, MA, USA; Elsevier: Amsterdam, The Netherlands, 2009; pp. 1–13. [Google Scholar]
  4. Perez, S.; Bertoft, E. The molecular structures of starch components and their contribution to the architecture of starch granules: A comprehensive review. Starch-Stärke 2010, 62, 389–420. [Google Scholar] [CrossRef]
  5. Xie, S.; Liu, Q.; Cui, S.W. Starch modification and applications. In Food Carbohydrates; Cui, S.W., Ed.; Taylor & Francis: Boca Raton, FL, USA, 2005; pp. 398–410. [Google Scholar]
  6. Keeling, P.L.; Myers, A.M. Biochemistry and genetics of starch synthesis. Annu. Rev. Food Sci. Technol. 2010, 1, 271–303. [Google Scholar] [CrossRef]
  7. Willet, J.L. Starch in polymer compositions. In Starch: Chemistry and Technology, 3rd ed.; BeMiller, J., Whistler, R., Eds.; Academic Press: Cambridge, MA, USA; Elsevier: Amsterdam, The Netherlands, 2009; pp. 715–743. [Google Scholar]
  8. Garimella-Purna, S.K.; Shi, Y.-C.; Guan, L.; Wilson, J.D.; Graybosch, R.A. Factors governing pasting properties of waxy wheat flours. Cereal Chem. 2015, 92, 529–535. [Google Scholar] [CrossRef]
  9. Concepcion, J.C.T.; Ouk, M.; Zhao, D.; Fitzgerald, M.A. The need for new tools and investment to improve the accuracy of selecting for grain quality in rice. Field Crops Res. 2015, 182, 60–67. [Google Scholar] [CrossRef]
  10. Batey, I.L. Interpretation of RVA curves. In The RVA Handbook; Crosbie, G.B., Ross, A.S., Eds.; AACC International: St. Paul, MN, USA, 2007; pp. 19–31. [Google Scholar]
  11. Wang, S.; Li, C.; Copeland, L.; Niu, Q.; Wang, S. Starch retrogradation: A comprehensive review. Compr. Rev. Food Sci. Food Saf. 2015, 14, 568–585. [Google Scholar] [CrossRef]
  12. BeMiller, J.N. Pasting, paste, and gel properties of starch-hydrocolloid combinations. Carbohydr. Polym. 2011, 86, 386–423. [Google Scholar] [CrossRef]
  13. Zhao, R.; Torley, P.; Halley, P.J. Emerging biodegradable materials: Starch- and protein-based bio-nanocomposites. J. Mater. Sci. 2008, 43, 3058–3071. [Google Scholar] [CrossRef]
  14. Peleg, M. The basics of solid foods rheology. In Food Texture; CRC Press: Boca Raton, FL, USA, 2017; pp. 3–33. [Google Scholar]
  15. Pojić, M.; Hadnadev, M.; Hadnadev, T.D. Gelatinization properties of wheat flour as determined by empirical and fundamental rheometric methods. Eur. Food Res. Technol. 2013, 237, 299–307. [Google Scholar] [CrossRef]
  16. Cozzolino, D.; Allder, K.; Roumeliotis, S.; Eglinton, J. Feasibility study on the use of multivariate data methods and derivatives to enhance information from barley flour and malt samples analysed using the Rapid Visco Analyser. J. Cereal Sci. 2012, 56, 610–614. [Google Scholar] [CrossRef]
  17. Cozzolino, D. The use of the rapid visco analyser (RVA) to sequentially study starch properties in commercial malting barley (Hordeum vulgare). J. Food Meas. Charact. 2016, 10, 474–479. [Google Scholar] [CrossRef]
  18. Balet, S.; Guelpa, A.; Fox, G.; Manley, M. Rapid Visco Analyser (RVA) as a tool for measuring starch-related physicochemical properties in cereals: A review. Food Anal. Methods 2019, 12, 2344–2360. [Google Scholar] [CrossRef]
  19. Abd Karim, A.; Norziah, M.; Seow, C. Methods for the study of starch retrogradation. Food Chem. 2000, 71, 9–36. [Google Scholar] [CrossRef]
  20. Oh, J.H.; Kim, M.J.; Yoo, B. Dynamic rheological properties of rice flour-starch blends. Starch-Stärke 2010, 62, 321–325. [Google Scholar] [CrossRef]
  21. McCleary, B.V.; Charmier, L.M.; McKie, V.A. Measurement of starch: Critical evaluation of current methodology. Starch-Stärke 2019, 71, 1800146. [Google Scholar] [CrossRef]
  22. Agelet, L.E.; Hurburgh, C.A. A tutorial on near-infrared spectroscopy and its calibration. Crit. Rev. Anal. Chem. 2010, 40, 246–260. [Google Scholar] [CrossRef]
  23. Szymańska, E.; Saccenti, E.; Smilde, A.K.; Westerhuis, J.A. Double-check: Validation of diagnostic statistics for PLS-DA models in metabolomics studies. Metabolomics 2012, 8, 3–16. [Google Scholar] [CrossRef]
  24. Szymańska, E.; Brodrick, E.; Williams, M.; Davies, A.N.; van Manen, H.-J.; Buydens, L.M.C. Data Size Reduction Strategy for the Classification of Breath and Air Samples Using Multicapillary Column-Ion Mobility Spectrometry. Anal. Chem. 2015, 87, 869–875. [Google Scholar] [CrossRef]
  25. Wongsaipun, S.; Krongchai, C.; Jakmunee, J.; Kittiwachana, S. Rice grain freshness measurement using Rapid Visco Analyser and chemometrics. Food Anal. Methods 2018, 11, 613–623. [Google Scholar] [CrossRef]
  26. Cozzolino, D.; Roumeliotis, S.; Eglinton, J. Combining partial least squares (PLS) discriminant analysis and Rapid Visco Analyser (RVA) to classify barley samples according to year of harvest and locality. Food Anal. Methods 2014, 7, 887–892. [Google Scholar] [CrossRef]
  27. Kaur, M.; Kaushal, P.; Sandhu, K.S. Studies on physicochemical and pasting properties of Taro (Colocasia esculenta L.) flour in comparison with cereal, tuber, and legume flour. J. Food Sci. Technol. 2013, 50, 94–100. [Google Scholar] [CrossRef] [PubMed]
  28. Zhu, L.; Wu, G.; Zhang, H.; Wang, L.; Qian, H.; Qi, X. Using RVA-full pattern fitting to develop rice viscosity fingerprints and improve type classification. J. Cereal Sci. 2018, 81, 1–7. [Google Scholar] [CrossRef]
  29. Deng, F.; Lu, H.; Yuan, Y.; Chen, H.; Li, Q.; Wang, L.; Tao, Y.; Zhou, W.; Cheng, H.; Chen, Y. Accurate prediction of the eating and cooking quality of rice using artificial neural networks and the texture properties of cooked rice. Food Chem. 2023, 407, 135176. [Google Scholar] [CrossRef]
  30. Fox, G.; Visser, J.; Skov, T.; Meijering, I.; Manley, M. Effect of different analysis conditions on Rapid Visco Analyser malt viscograms in relation to malt of varying fermentability. J. Inst. Brew. 2014, 120, 183–192. [Google Scholar] [CrossRef]
  31. Kim, H.; Jeong, S.; Kwak, J.; Lee, S. Artificial intelligence classification and amylose content prediction of rice flour varieties from their pasting features. J. Agric. Food Res. 2023, 14, 100847. [Google Scholar] [CrossRef]
  32. Ohm, J.B.; Ross, A.; Ong, Y.L.; Peterson, C. Using multivariate techniques to predict wheat flour dough and noodle characteristics from size-exclusion HPLC and RVA data. Cereal Chem. 2006, 83, 1–9. [Google Scholar] [CrossRef]
  33. Saleh, M.; Amr, A.; Mehyar, G.; Ondier, G. Predicting farinograph parameters by Rapid Visco Analyser pasting profile using partial least square regression. Qual. Assur. Saf. Crops Foods 2016, 8, 41–49. [Google Scholar] [CrossRef]
  34. Sampaio, P.S.; Almeida, A.S.; Brites, C.M. Use of artificial neural network model for rice quality prediction based on grain physical parameters. Foods 2021, 10, 3016. [Google Scholar] [CrossRef]
  35. Nastasi, J.R.; Ma, S.; Alagappan, S.; Hoffman, L.C.; Cozzolino, D. Assessment of the addition of cricket (Acheta domesticus) powder to chickpea (Cicer arietinum) and flaxseed (Linum usitatissimum) flours: A chemometric evaluation of their pasting properties. Appl. Sci. 2024, 14, 7131. [Google Scholar] [CrossRef]
  36. Shafie, B.; Cheng, S.; Lee, H.; Yiu, P. Characterization and classification of whole-grain rice based on Rapid Visco Analyser (RVA) pasting profile. Int. Food Res. J. 2016, 23, 5. [Google Scholar]
  37. Meadows, F. Pasting process in rice flour using Rapid Visco Analyser curves and first derivatives. Cereal Chem. 2002, 79, 559–562. [Google Scholar] [CrossRef]
  38. Juhasz, R.; Salgo, A. Pasting behaviour of amylose, amylopectin and their mixtures determined by RVA curves and first derivatives. Starch-Stärke 2008, 60, 70–78. [Google Scholar] [CrossRef]
  39. Guelpa, A.; Bevilacqua, M.; Marini, F.; O’Kennedy, K.; Geladi, P.; Manley, M. Application of Rapid Visco Analyser (RVA) viscograms and chemometrics for maize hardness characterisation. Food Chem. 2015, 173, 1220–1227. [Google Scholar] [CrossRef]
  40. Mijland, P.J.; Janssen, A.M.; de Vries, P.A. Multivariate comparison of the Brabender Viscograph and the Rapid Visco Analyser using cross-linked starch. Starch-Stärke 1999, 51, 33–39. [Google Scholar] [CrossRef]
  41. Buenafe, R.J.; Rathnam, A.; Añonuevo, J.J.; Sundar, S.; Sreenivasulu, N. Application of classification models in screening superior rice grain quality in male sterile and pollen parents. J. Food Compos. Anal. 2021, 104, 104137. [Google Scholar] [CrossRef]
  42. Dayananda, B.; Owen, S.; Kolobaric, A.; Chapman, J.; Cozzolino, D. Pre-processing applied to instrumental data in analytical chemistry: A brief review of the methods and examples. Crit. Rev. Anal. Chem. 2024, 54, 2745–2753. [Google Scholar] [CrossRef]
  43. Serna-Saldivar, S.O.; Jacobo-Velázquez, D.A. Quality Control and Assurance in the Snack Food Industry. In Snack Foods; CRC Press: Boca Raton, FL, USA, 2022; pp. 477–518. [Google Scholar]
  44. Rose, K.; Eldridge, S.; Chapin, L. The Internet of Things: An Overview. Internet Soc. (ISOC) 2015, 80, 1–53. [Google Scholar]
  45. Rahul, K.; Banyal, R.K.; Kumar, V.; Hooda, D. Cloud Computing: Technological Innovations in the Food Industry. In Operations and Supply Chain Management in the Food Industry: Farm to Fork; Springer Nature: Singapore, 2022; pp. 127–141. [Google Scholar]
  46. Chhetri, K.B. Applications of Artificial Intelligence and Machine Learning in Food Quality Control and Safety Assessment. Food Eng. Rev. 2024, 16, 1–21. [Google Scholar] [CrossRef]
  47. Perten Science World. Issue No 11. Available online: https://www.perten.com/Global/PSW/PSW_No11.pdf (accessed on 4 March 2025).
Applsci 15 03376 g001
Figure 1. Typical Rapid Visco Analyser (RVA) profile of different rice flours with varying amylose contents, indicating the main parameters that are calculated during the analysis.
Figure 1. Typical Rapid Visco Analyser (RVA) profile of different rice flours with varying amylose contents, indicating the main parameters that are calculated during the analysis.
Applsci 15 03376 g001
Table 1. Rapid Visco Analyser (RVA) parameters, common abbreviations, definitions, and analytical significance.
Table 1. Rapid Visco Analyser (RVA) parameters, common abbreviations, definitions, and analytical significance.
ParameterAbvs.DefinitionSignificance
Pasting TemperaturePT, PTempThe temperature at which the sample begins to thicken, and the viscosity starts to rise. It marks the onset of starch gelatinization as the granules absorb water, swell, and disrupt.PT indicates the gelatinization temperature, which is important for understanding the processing conditions required for starch-based products.
Peak ViscosityPVThe highest viscosity reached during the heating phase of the RVA analysis. It occurs when starch granules are fully swollen and before they start to break down.PV reflects the maximum water-holding capacity and swelling of the starch granules. It is crucial for determining the thickening potential of starch in a product.
Time to PeakPTi, TTP, TpeakThe time it takes to reach the peak viscosity during the analysis.PTi provides information on how quickly the starch reaches its maximum viscosity, which is useful for evaluating processing times.
BreakdownBD, BVis, BKDThe difference between the peak viscosity (PV) and Trough/Minimum Viscosity (TV). It represents the reduction in viscosity due to the breakdown of starch granules under heat and shear.BD, or BKD, indicates the stability of starch under heat and shear stress. A higher breakdown means that the starch is less stable and more prone to degradation, which can affect the texture and viscosity in food products.
Trough/Minimum ViscosityTV, Tmin, Trough, THThe lowest viscosity after peak viscosity during the holding period, representing the point at which the starch granules are most disrupted.TV, or TH, gives insight into the extent of granule breakdown and the ability of the starch to maintain viscosity during processing.
SetbackSB, STBThe difference between final viscosity (FV) and Trough Viscosity (TV). It measures the increase in viscosity during the cooling phase as the starch granules start to reassociate and form a gel.SB, or STB, is important for understanding starch retrogradation, which impacts the texture, firmness, and shelf life of starch-based products.
Final ViscosityFVThe viscosity at the end of the RVA test during the cooling phase. It reflects the viscosity of the starch gel after it has been heated and then cooled.FV is a key indicator of the final texture and stability of starch-based products, such as sauces, gels, and baked goods. It also provides insight into starch retrogradation.
Table 2. Summary of applications of the Rapid Visco Analyser (RVA) combined with machine learning tools as reported in the scientific literature.
Table 2. Summary of applications of the Rapid Visco Analyser (RVA) combined with machine learning tools as reported in the scientific literature.
Sample, Crop, or FoodMachine Learning ToolSee References for the Specific Application
Maize, Rice, Barley, Cricket, Flaxseed, Chickpea, Wheat, Cassava, Tapioca, Proso Millet, Starch PCA[16,17,25,32,36,39,40]
Barley, Rice, Starch, Cricket, Flaxseed, ChickpeaPLS/PLS-DA[16,17,25,33,34,35,40]
RiceSVM[31,41]
RiceANN[29,34]
RiceKNN[29,41]
ANN: artificial neural network; PCA: principal component analysis; PLS: partial least squares; PLS-DA: partial least squares discriminant analysis; SVM: support vector machines.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Nastasi, J.R.; Alagappan, S.; Cozzolino, D. The Combination of Machine Learning Tools with the Rapid Visco Analyser (RVA) to Enhance the Analysis of Starchy Food Ingredients and Products. Appl. Sci. 2025, 15, 3376. https://doi.org/10.3390/app15063376

AMA Style

Nastasi JR, Alagappan S, Cozzolino D. The Combination of Machine Learning Tools with the Rapid Visco Analyser (RVA) to Enhance the Analysis of Starchy Food Ingredients and Products. Applied Sciences. 2025; 15(6):3376. https://doi.org/10.3390/app15063376

Chicago/Turabian Style

Nastasi, Joseph Robert, Shanmugam Alagappan, and Daniel Cozzolino. 2025. "The Combination of Machine Learning Tools with the Rapid Visco Analyser (RVA) to Enhance the Analysis of Starchy Food Ingredients and Products" Applied Sciences 15, no. 6: 3376. https://doi.org/10.3390/app15063376

APA Style

Nastasi, J. R., Alagappan, S., & Cozzolino, D. (2025). The Combination of Machine Learning Tools with the Rapid Visco Analyser (RVA) to Enhance the Analysis of Starchy Food Ingredients and Products. Applied Sciences, 15(6), 3376. https://doi.org/10.3390/app15063376

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop