Machine learning based analysis and prediction of crop yield and prices of Aman, Aus and Boro rice

Abstract

Agriculture has been the driving force of the Bangladesh economy. In the agricultural sector, farmers are largely incapable of using scientific technology to maximize crop yield and identify which crops can be grown in specific weather and soil conditions. Recently, the effectiveness of machine learning-based algorithms in utilizing large datasets to accurately predict and provide descriptive solutions holds promising potential in solving this problem by giving descriptive farming advice and fertilizer usage for farmers and proper yield predictions for better import and export policies. Therefore, this paper aims to use historical weather and climate data (such as temperature, rainfall, average bright sunshine, cloud coverage, etc.) and agricultural data such as fertilizer, soil type, and soil moisture to provide predictions on the yield of Aus, Boro, and Aman that can be expected to grow in a region as well as predict the future rice prices of Dhaka depending on existing data. After analysis it was found that there is direct correlation of high accuracy between weather factors such as average rainfall, average minimum temperature, average maximum temperature, average yearly temperature, average bright sunshine, average cloud coverage, relative humidity, average wind speed, latitude, longitude and altitude and yearly yield of Aus, Aman and Boro rice when algorithms such as KNN, linear regression, random forest, and XGBoost were implemented. Furthermore, correlation was found among soil type, soil moisture, fertilizer type and crop yield. Finally, a price prediction of three different types of rice –Aus, Aman, and Boro – between Dhaka and Delhi was conducted using models such as ARIMA and SARIMAX.