Exogenous variable driven deep learning models for improved price forecasting of TOP crops in India.

Area of Science:

• Agricultural Economics
• Data Science
• Time Series Forecasting

Background:

• Accurate agricultural commodity price prediction is vital for India's economy.
• Traditional models face limitations with assumptions and feature extraction.
• Existing machine learning (ML) and deep learning (DL) approaches often overlook crucial exogenous factors.

Purpose of the Study:

• To evaluate advanced deep learning models (NBEATSX, TransformerX) for agricultural price forecasting in India.
• To investigate the impact of incorporating exogenous variables, specifically weather data, into DL models.
• To compare the performance of DL models against traditional statistical and ML methods.

Main Methods:

• Utilized price data for Tomato, Onion, and Potato (TOP) from major Indian markets.
• Integrated corresponding weather data (precipitation, temperature) as exogenous variables.
• Compared NBEATSX and TransformerX against ARIMAX, MLR, ANN, SVR, RFR, and XGBoost, evaluating performance using RMSE, MAE, sMAPE, MASE, and QL metrics.

Main Results:

• Deep learning models, NBEATSX and TransformerX, demonstrated superior performance.
• These DL models, especially when augmented with weather data, consistently outperformed statistical and ML models.
• NBEATSX and TransformerX achieved average RMSEs of 110.33 and 135.33, respectively, with other error metrics also showing significant improvements.

Conclusions:

• Advanced deep learning models, NBEATSX and TransformerX, are highly effective for agricultural commodity price forecasting in India.
• Incorporating exogenous variables like weather data significantly enhances the predictive power of DL models.
• This research highlights the potential of DL for more accurate and reliable agricultural price predictions, addressing a key research gap.