Advanced Potato Price Prediction through N-BEATS Deep Learning Architecture

Abstract

Agricultural commodity prices exhibit unique challenges due to seasonality, inelastic demand, and production uncertainty, leading to significant fluctuations in time series data. This paper explores these complexities by applying Deep Learning (DL) models to forecast agricultural prices, specifically focusing on potato prices. While DL models have excelled in domains like image processing and natural language processing, they require specialized architectures for effective time series forecasting. This study evaluates the Neural Basis Expansion Analysis for Interpretable Time Series Forecasting (N-BEATS) model, a novel DL architecture designed for time series data using daily potato price data from the Azadpur market in Delhi, spanning January 1, 2018, to April 30, 2023.The performance of N-BEATS is compared with three baseline models: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU). Evaluation criteria include Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The results show that the N-BEATS model consistently outperforms the other models across all metrics. Additionally, the Diebold-Mariano (DM) test confirms the N-BEATS model's superior forecasting accuracy compared to the other models. This research highlights the potential of the N-BEATS model to significantly enhance the precision of agricultural price forecasting, providing valuable insights for farmers, planners, and other stakeholders in the agricultural sector.