Research on vegetable automated operation decision-making based on data mining methods using random forest, LSTM-Seq2Seq, and NSGA-II

Vegetables, characterized by short shelf life and easily perishable features, pose operational challenges for fresh food supermarkets in replenishing and pricing vegetable products within limited sales space. This study proposes an automated decision-making framework based on machine learning, data mining, and intelligent optimization algorithms, using a random forest model for regression prediction and an LSTM-based Seq2Seq model for time series prediction to address these challenges. The random forest model is trained using historical wholesale price data for regression prediction, with an average root mean square error (RMSE) of 0.7046 on the test set. In the time series prediction based on historical sales data, an LSTM-based Seq2Seq model is used, resulting in an average RMSE of 0.9631 on the test set. Subsequently, the random forest model is trained to predict actual sales volume based on the relationship between sales volume, wholesale price, and sales price, combining it with a multi-objective optimization model. The NSGA-II intelligent optimization algorithm is employed to solve the model. Ultimately, we determined various vegetables' sales prices and replenishment quantities by maximizing profits and minimizing supply surplus losses. The method proposed in this study helps improve supermarkets' operational efficiency and profits, providing new insights and methods for fresh food management decision-making