Prediction of retail commodity hot-spots: a machine learning approach

The accurate prediction of hot-spot commodities in the retail market is crucial for inventory management and market strategy formulation. Traditional methods focus on analyzing the sales trends of listed commodities to mine hot-spot commodities; however, these methods are incapable of dealing with new commodities that lack historical sales data. Although existing studies have used commodity features to predict hot-spot commodities, commodity features are complex and diverse, and it is difficult for a single prediction model to fully capture and accurately analyze the relationship between commodity features and sales hot-spots. This study proposes a commodity hot-spot classification prediction method based on an ensemble learning model that combines multiple machine learning methods, namely the gradient boosting decision tree, eXtreme gradient boosting, and light gradient-boosting machine, to analyze the relationship between commodity features and retail market sales trends. In addition, a voting strategy is designed to fuse the model classification results. The experimental results show that the accuracy of the ensemble model is as high as 0.91, outperforming a single model and other approaches. This method can effectively help enterprises to identify hot-spot commodities and provide strong support for new commodity designs and sales strategy adjustments.