An advanced wide-and-deep learning framework for soybean price forecasting using market, weather, trade, and supply data.

Precise prediction of agricultural commodity prices is key to assisting farmers, traders, policymakers and supply chain stakeholders to alleviate the uncertainty, and enable them to make sound decisions. The fluctuation in the price of soybean is very unpredictable because of the interplay of market forces, weather unpredictability, trade, output and farming methods. This paper suggests a multi-source machine learning model that is capable of forecasting the accurate price of soy beans using past market prices, arrivals, trade as well as weather forecasting, crop supply and area under cultivation. The dataset of a complete prediction of the prices of soybean between Jan 2015 to March 2026 was built by using the data provided in AGMARKNET, NCDEX, government trade portals, meteorological agencies and state agricultural databases. Exploratory data analysis was done to learn price distributions and regional differences, then systematic data preprocessing, normalization and feature engineering such as rainfall deviation, price range and arrivals per cultivation area was done. Several machine learning and deep learning models were implemented, such as Random Forest, Gradient Boosting, XGBoost, ensemble methods, dense neural network, attention-based models, and a proposed Wide and Deep Network (AgroWDN) was tested. RMSE, MAE, MSE and R