Platform Merchant Demand Prediction Based on Decision Tree and Multi-Layer Perceptron Models

Under the background of platform merchant multidimensional demand prediction, there is the rise of the e-commerce industry and the development of the logistics industry, the need to accurately predict the supply chain demand, e-commerce platforms can effectively arrange logistics, inventory and other resources to improve operational efficiency and meet consumer demand. In this paper, we need to use historical data to predict the future demand for products using a time-series feature-based prediction model. Data preprocessing is performed first. In the next step, this paper trains four regression models, XGBoost [1], Random Forest [2], Decision Tree [3] [4], and Multi-Layer Perceptron Regression [5] [6], by using the first 90% of the data in the data table, and selects Decision Tree as the optimal regression model using the second 10% of the data in the data table, and the model has a better overall prediction effect using the 1-wmape as the criterion. Finally, this paper predicts the demand for each merchant's goods in each warehouse from 2023-05-16 to 2023-05-30. For the classification of time series formed by merchants, warehouses, and commodities, this paper determines the optimal number of clusters as 5 classes using the K-means clustering algorithm, and the same class of time series data has the most similar characteristics in terms of demand.