A Hybrid Temporal Convolutional Network and Transformer Model for Accurate and Scalable Sales Forecasting

Abstract

Accurate product sales forecasting is critical for inventory management, pricing strategies, and supply chain optimization in the retail industry. This article proposes a novel deep learning architecture that integrates Temporal Convolutional Networks (TCNs) with Transformer-based attention mechanisms to capture both short-term and long-term dependencies in time-series sales data. Utilizing the Favorita Grocery Sales Forecasting dataset, our hybrid TCN Transformer model demonstrates superior performance over existing models by incorporating external factors such as holidays, promotions, oil prices, and transaction data. The model achieves state-of-the-art results with a Mean Absolute Error (MAE) of 2.01, Root Mean Squared Error (RMSE) of 2.81, and a Weighted Mean Absolute Percentage Error (wMAPE) of 4.22%, significantly outperforming other leading models such as LSTM, GRU, and TFT. Extensive cross-validation confirms the robustness of our model, achieving consistently high performance across multiple folds.