Machine Learning-based Crop Yield Forecasting in Bangladesh: An Early Prediction Approach.
No Thumbnail Available
Files
Date
2024-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
International Islamic University Chittagong
Abstract
Farmers in Bangladesh are facing substantial financial losses due to challenges in selecting the optimal crops for production, influenced by factors such as pesticides, fertilizer, and annual rainfall. The agriculture supply chain can benefit from yield prediction to make crucial decisions that mitigate risks associated with changes in crop growth. This research focuses on the critical task of early crop yield prediction in Bangladesh, specifically targeting rice, sugarcane, potato, and maize. A diverse set of machine learning algorithms, including K-Nearest Neighbors, Random Forest, Gradient Boosting, AdaBoost, CatBoost, Decision Tree, and XGBoost, are employed in the initial phase to assess their individual predictive performances. After a comprehensive analysis, K-Nearest Neighbors, Random Forest, and Gradient Boosting emerge as the three most effective models. The study progresses to explore ensemble learning techniques, utilizing a Voting Regression in these top three models to combine their predictions. Furthermore, advanced ensemble methodologies, specifically stacking with varying both final and base estimators using the three best models, are implemented to harness the collective intelligence of diverse models and enhance overall predictive accuracy. The methodology includes a rigorous evaluation process, considering various metrics to assess the effectiveness of the models and ensemble techniques. The findings provide valuable insights into how different algorithms and ensemble methodologies can collaborate to optimize early crop yield prediction for specific crops in Bangladesh. Among all combinations, for maize crop yield prediction achieved R2 0.99 and MAE 0.04, rice crop yield prediction achieved R2 0.99 and MAE 0.06, potato crop yield prediction achieved R2 0.99 and MAE 0.04, sugarcane crop yield prediction achieved R2 0.99 and MAE 0.36. This research contributes not only to the field of agricultural prediction but also to the broader application of ensemble learning in optimizing machine learning models for complex tasks. The findings are significant for stakeholders in agriculture, providing them with reliable tools for early crop yield estimation, crucial for effective resource management and decision-making.
Description
This internship report is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science in B.Sc. Engineering, 2024
Cataloged from the PDF version of the internship report