Harvesting Insights: AI-Driven Rice Yield Predictions and Big Data Analytics in Agriculture ^†

Abstract

This paper explores the transformative potential of artificial intelligence (AI) and big data analytics in predicting rice yields within the agricultural domain. By employing advanced algorithms and innovative methodologies, our study aims to contribute to the optimization of crop management strategies, providing a glimpse into the future of sustainable agriculture. The integration of AI and big data analytics allows us to unveil novel insights into rice yield predictions, emphasizing their broader implications for global food security. Our optimized Random Forest Regression model exhibited impressive results, with a Mean Forecasting Error (MFE) of 0.0001, a Mean Absolute Error (MAE) of 0.00016, a Mean Square Error (MSE) of 0.000014, and a Root-Mean-Square Error (RMSE) of 0.003. Our innovative methodologies involve combining climatic data, rice yield from previous seasons, and cultivated areas as input variables for prediction models. Additionally, we employ advanced optimization methods such as Optuna and Hyperopt to enhance our model. The integration of AI with big data analytics into rice yield predictions aids in preparing the data to achieve high quality before applying our models. This includes selecting optimal features and simulating our model with generated data to ensure it avoids overfitting. The potential of our approach has led to the creation of a digital agricultural twin for monitoring, analyzing, and visualizing data provided by sensors installed on a farm in Rosso during the study period. Looking forward, this digital twin enhances precision agriculture practices, contributing to sustainable farming and global food security. As a future perspective, we aim to create an intelligent system using our models and integrate IoT technologies to expose our model results as a service. We plan to publish the first Mauritanian agricultural database for other researchers to use in their future research.