Predictors selection strategy based on stepwise random forests and logistic regression model
Chaozhi Li · 2023
Random Forests (RF) are a popular machine learning method for developing variables selection models. However, it may suffer from a lack of better model interpretability compared with traditional models, such as Logistic Regression (LR). In this paper, we propose a predictors selection strategy based on Stepwise Random Forests and Logistic Regression model (SRFLR) and validate it using HCV data provided by UCI machine learning repository platform. The Synthetic Minority Oversampling Technique (SMOTE) algorithm is adopted to deal with the problem of imbalance class in the dataset. The results demonstrate that the proposed SRFLR can obtain more predictors while preserving better prediction ability than LR alone, which will offer some references for clinical researchers to select relevant disease predictors.