Machine Learning-Driven Approch of Crop Selection for Sustainable Agriculture in Bangladesh

In Bangladesh, agriculture serves as a cornerstone of the economy and a vital source of livelihood for millions. Optimizing crop selection is paramount for enhancing agricultural productivity and ensuring food security. This paper presents a Decision Support System DSS that would suggest the most optimal crops to farmers after conducting an in-depth analysis of critical environmental factors. The system uses a large dataset containing records for 40 of the most harvested crops in Bangladesh, along with required parameter profiles on average monthly rainfall (in mm), temperature (°C), humidity (%), sunlight duration (in hours), wind direction (°), and wind velocity (km/h) coupled with temporal and geographical variables. Now, these data have been screened through different machine learning models to look for the most appropriate crop choices for future harvests. The most proficient among the evaluated classifiers was Random Forest with an accuracy of 98.11%, a weighted precision of 98.12%, recall of 98.11%, and an F1 score of 98.11%. From the rest notable models include Decision Tree, CNN, XGBoost. An ensemble approach of the top five classifiers also performed well, with an accuracy of 0.9648. The DSS will enable farmers not only with insights gained from data but also promote environmentally viable agricultural practices by matching crop selection with prevailing conditions. This study outlines the transformational impact of machine learning on decision-making in agriculture, where recommendations would indeed be sensitive to specific environmental variables to make sure the best possible yields and improved food security in Bangladesh.