For the e-commerce platform, obtaining sales status of the stores is important for formulating sale strategies, risk assessment plans and loan appraisal. The traditional way to obtain sales status is mainly based on the subjective judgment of relevant practitioners and the analysis of mathematical statistical models composed of historical data. These methods are inaccurate and too dependent on people’s judgment. Therefore, using data mining and machine learning technology to predict the sales amount came into being. In this paper, we propose a method to process a great deal of data from China's famous e-commerce platform called Jingdong. This method can make the messy data become uniform data sets which are more suitable for machine learning. Based on the uniform data sets, two sales prediction models are used to predict the sales amount of the stores in Jingdong. In experiment, 9-month historical sales and behavior data of 10,000 stores in Jingdong platform are processed by the proposed method. Furthermore, two prediction models including GBDT(Gradient Boosting Decision Tree)+DNN(Deep neural network) and GA(genetic algorithm) are used to predict the sales amount of stores in following 3 months. To verify the accuracy of the prediction, we import WMAE?Weighted Mean Average Error?score. In experimental results, the best WMAE is 0.39, which means accuracy is 61%. It shows the method of data processing and prediction models are effective compare with other models. This indicates the proposed method and model can be used for sales prediction in e-commerce platform.