Productivity Analysis of Potato (Solanum tuberosum) Yielding in Cumilla, Bangladesh Based on Soil Chemical Parameters Using Machine Learning Approaches.

Abstract

The aim of this project is to apply machine learning methods to evaluate the productivity of potato yields in Cumilla, Bangladesh, based on soil chemical parameters. Collaborating with the Regional Agricultural Research Station, Bangladesh Agriculture Research Institute (BARI) in Cumilla, a comprehensive dataset spanning crop yields, soil quality, climate conditions, and region-specific agronomic practices has been developed. This dataset was subjected to 277 machine learning classifiers in order to determine the most effective techniques for predicting potato productivity. Based to the analysis, the top 25 classifiers including AdaBoost, GLMBoost, CNN, Simulated Annealing, Bayesian Optimization, GAN, Multi-Layer Perceptron, Support Vector Machine, FDA, Deep-Q-Network, Linear Regression, LDA, Ensemble Model, Autoencoder provided significant insights into the parameters influencing potato yield, with soil chemical characteristics emerging as key influences. Simulated Annealing, Bayesian Optimization, MLP, and CNN exhibit outstanding results with 90%, 89%, 88%, and 87% accuracy, respectively. despite this, GAN approaches and become the best match for the dataset with 99.96% accuracy. The correctness and applicability of the dataset to the regional agricultural environment were validated during the validation procedure. The relationship between crop yield and soil properties has become clearer because of to these discoveries, which have real-world implications for enhancing agricultural practices in Bangladesh. The verified dataset supports data-driven decision-making and sustainable development objectives by bridging a critical gap in the local agricultural research infrastructure