Comparative Performance Analysis of ARIMA and Time Delay Neural Networks for Cauliflower Price Forecasting in North India
Price: ₹ 500
Author: Divyanshu, Chandresh Guleria, Pardeep Singh, Subhash Sharma, Rohit Kumar Vashishat and Priyank Sharma
Author Address: enior Research Fellow, Assistant Professor, and 4Associate Professor and Head, Agricultural Economics, Department of Social Sciences, Dr. YSP UHF Nauni, Solan-173 230 (H.P), Department of Social Sciences, Dr. YSP UHF Nauni, Solan (H.P)-173 230, Research A
Keywords: Box Jenkins methodology, forecasting, neural networks, root mean square error, time series.
JEL Codes: C10, C52, C53, Q11.
Abstract
Accurate price forecasting of perishable crops is crucial for farmers' income stability. This study compared ARIMA and TDNN models across selected markets. The results indicated that ARIMA performed best in Ludhiana with the {(0,1,1) (0,1,0)} model, yielding the lowest RMSE (103.22). However, TDNN outperformed ARIMA in most cases, with Ludhiana again showing the best performance (RMSE 161.77) using the (2:5s:1l) neural network model. These findings highlighted TDNN's superiority in capturing non-stationary price patterns. Policymakers should integrate TDNN-based forecasting into agricultural price intelligence systems to improve market predictability. Reliable price forecasts can aid farmers in better planning, reducing distress sales, and enhancing profitability. Additionally, government initiatives should promote digital tools and machine learning models for real-time price predictions. Strengthening price forecasting mechanisms will help mitigate market volatility, support procurement decisions, and enhance overall agricultural sustainability in India.
Description
Indian J Econ Dev, 2025, 21(1), 48-60
https://doi.org/10.35716/IJED-23529