DEVELOPING MACHINE LEARNING MODEL FOR ACCURATE PREDICTION OF SOYBEAN MARKET PRICES USING MULTIDISCIPLINARY AGRICULTURAL DATA SOURCES

Abstract

Accurately predicting the prices of farming commodities is a key part of keeping markets stable and helping farmers make decisions.  Using both machine learning and deep learning, this study tries to guess what the prices of soyabean, one of India's main cash crops, will be in the future.  The data for this study came from a number of reliable sources, including AGMARKNET, NCDEX, the Department of Agriculture (Maharashtra State), IMD Pune, and commerce.gov.in. It was put together by the researchers themselves and covers the years 2015–2025.  Different factors that affect price changes were looked at, such as changes in weather, trade data, crop landings, farming area, and import-export trends. A lot of work went into cleaning the data, like fixing missing values, normalising it, and adding new features (like rainfall difference, visits per area, and price range). We used and compared a number of different prediction models, such as Random Forest, Gradient Boosting, XGBoost, Ensemble methods, and a number of deep learning designs, such as Dense Neural Network (DNN), Attention Model, Regularised Network, and an AgroWDN that was specifically built for this study.  Metrics like RMSE, MAE, MSE, and R² were used to judge performance. AgroWDN did the best out of all the models, with an RMSE of 111.43, an MAE of 57.15, and a R² of 98.96%, doing better than traditional machine learning methods.  The results show that complicated agro-economic data can effectively capture both linear and nonlinear relationships when built with mixed designs that combine wide and deep learning.  This research shows a strong method for predicting farming prices that can help lawmakers, buyers, and farmers make the best decisions about how to plan crops and handle risks.