At the current stage of rapid popularization of electric vehicles in our country, the forecast of electric vehicle ownership has attracted widespread attention. This paper proposed an adaptive promotion method based on mutual information feature selection to carry out research on the prediction of electric vehicles ownership. Firstly, the data set is interpolated as well as data-partitioned, and the rationality of the input variables is analyzed. Secondly, the same test input variables output variables ownership degree of mutual information, and the dependent variable to be weaker excluded. Finally, the matrix input adaptive boosting algorithm variable regression model by the mutual information feature selection, based on the obtained electric vehicle MIFS-AdaBoost model predicted value of ownership. Using MAE, MSE and $\mathrm{R}^{2}$ three regression evaluation indicators to compare the prediction effect of the model in this paper with seven types of machine learning regression models, the results show that the model proposed in this paper has the smallest prediction error and the highest coefficient of determination, which is better than other regression models.