Research on sales strategy of electric vehicle target customers based on machine learning algorithm

Abstract

Due to the influence of various subjective and objective factors, there is great uncertainty in consumers' willingness to purchase new energy vehicles. In this study, it is hoped that based on customer satisfaction and personal characteristic information, the main factors affecting whether the target customers are willing to purchase new energy vehicles are explored, so that corresponding sales strategies can be formulated. In this study, two different embedding methods based on penalty terms and tree models are used for feature selection, the former using three models LR, LASSO and SVM, and the latter using RF and LightGBM models for a total of five models for machine learning to find out the relevant features that affect the sales of different brands, and the voting method is applied to the selected features, and the results are found. The battery technology performance of electric vehicles, comfort, annual mortgage of target customers and the ratio of auto loan to annual household income have a significant impact on the sales of all three brands. In addition, affordability, safety and customer's work situation also had different degrees of influence on sales of the different brands. A multilayer perceptron (BP neural network) prediction model is built for each brand's target customers to predict their purchase intention. The prediction accuracy of the network is improved by balancing the positive and negative samples so that the ratio of the two types of samples in the dataset is approximately 1:1. The test set is also used for real-time testing to determine the fit status of the dataset. The accuracy of the final model was above 80% on both the training and test sets, and the model predicted the 1st, 5th, 6th, 7th, 12th, and 12th samples. The model predicted the purchase intention of the 1st, 5th, 6th, 7th, 12th and 13th customers.