Forecasting Jute Prices and Arrivals in West Bengal using ARIMA and advanced Machine Learning techniques

Abstract

This study aims to forecast and understand the price movements and volatility in jute prices and arrivals across the markets of four major jute growing districts in West Bengal namely Uttar Dinajpur, Nadia, Coochbehar, and Murshidabad. To achieve this, both traditional statistical methods (ARIMA) and advanced machine learning models, including Artificial Neural Networks (ANN), Support Vector Machine (SVM), Random Forest (RF), and Long Short-Term Memory (LSTM), were employed. The model performance was evaluated using error metrics such as Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). The LSTM model outperformed the others, with an average RMSE of 12.53 and a MAPE of 4.87%, demonstrating its superior accuracy in forecasting jute prices. In the case of jute arrivals, LSTM and SVR achieved the best performance, with LSTM recording the lowest RMSE of 14.32 and MAE of 11.87 in predicting arrivals. The findings underscore the non-linearity and interdependencies between markets, providing critical insights for traders and policymakers. These insights enable more precise anticipation of price fluctuations and contribute to better-informed market strategies, ultimately benefiting the jute supply chain in West Bengal.