Convolutional Neural Networks for Advanced Sales Forecasting in Dynamic Market Environments

This paper presents an enhanced approach to sales forecasting using advanced hybrid deep learning architectures, specifically Convolutional Neural Networks (CNNs) combined with Residual Networks (ResNets) and Temporal Convolutional Networks (TCNs). Utilizing the “Store Item Demand Forecasting Challenge" dataset, the study demonstrates significant improvements in forecasting accuracy over traditional models. The enhanced CNN-TCN model achieved the lowest Mean Absolute Percentage Error (MAPE) of 2.0% and the highest Prediction Interval Coverage Probability (PICP) of 96%. These results highlight the potential of hybrid architectures to provide more reliable and precise sales forecasts, offering valuable insights for business decision-making and strategic planning.