Edible Oil Price Forecasting: A Novel Approach with Group Temporal Convolutional Network and BetaAdaptiveAdam

Edible oil, a fundamental food commodity, plays a pivotal role in the economic progression of a nation. The precise forecasting of edible oil prices is of utmost significance to a broad spectrum of stakeholders, including investors, policymakers, and researchers. Over recent years, a myriad of factors, including international influences, have led to substantial price fluctuations and irregular cycles in edible oil markets. As a result, the task of achieving accurate and robust forecasts of edible oil prices has emerged as a daunting challenge. In response to this challenge, this study introduces a novel forecasting framework specifically designed for the prediction of Chinese edible oil prices. This framework incorporates Group Temporal Convolutional Networks (GTCN) and BetaAdaptiveAdam. The proposed methodology employs Singular Spectrum Analysis (SSA) to decompose the original data into multiple subseries. It then utilizes GTCN to extract temporal features from each subseries and integrates BetaAdaptiveAdam to enhance model generalization. The framework subsequently generates predictions for each subseries and ultimately amalgamates the results of each component to produce the final price predictions for the original series. Extensive experiments have been conducted to validate the proposed forecasting framework. The results underscore the exceptional performance of the proposed framework in terms of regression accuracy and directional prediction precision.