Improved Logistic Regression Model Based on Resampling Techniques

Abstract

When facing imbalanced samples and high-dimensional features, the performance of traditional logistic regression models may significantly deteriorate, even becoming completely ineffective. Therefore, this paper proposes an improved logistic regression method combined with resampling techniques. Firstly, the proposed method uses the "resampling and encoding" strategy to effectively capture the predictive information that has a significant impact on the classification result while addressing the problems of imbalanced samples and dimension curse. Secondly, the proposed method uses weighted combination of multiple submodels to further use the logistic regression model, thereby improving the model's adaptability and robustness. Furthermore, considering that the number of submodels required to cover all features may be excessive, this paper introduces L1 regularization to penalize each logistic regression submodel, further alleviating the impact of dimension curse on the prediction model. Simulation experiments and real-data analysis show that the proposed method has significantly improved prediction accuracy compared to traditional logistic regression, and has demonstrated certain competitiveness in comparison with some classic classification models.