In the era of rapid development of information technology, the e-commerce industry is also booming. The sales problem greatly affects the merchants’ sales revenue and the determination of marketing programs, so it is crucial for merchants to grasp the dynamics of the sales of goods in a timely manner. This study explores the e-commerce agricultural products feature data set and data mining by constructing it. Taking this data as the data base, a single model-based demand prediction model is constructed. And train ARIMA, LSTM, Random Forest models, are trained to solve the problem that the artificial fixed parameter of machine learning model lacks reasonableness, LSTM and Random Forest were optimized and improved first by using PSO and Bayesian algorithm respectively, and then selecting the optimal parameters to constitute prediction model.