Robust Variable Selection And Application In Logistic Regression Model

With the help of machine learning algorithm,we can mine the properties which contained in large dataset more effectively than before,which is of great significance to solve practical problems.Especially in biomedical fields,logistic model classifier has been widely used.However,this classification method is very sensitive to contaminated data.When there are outliers in the dataset,the logistic regression estimator will have serious impact and bias.Aiming at the robustness of logistic regression model,a robust penalty regression method with adaptive weights based on MM algorithm and Fast-MCD estimation theory is proposed,and its performance of robustness and effectiveness is evaluated via numerical simulations.Finally,the proposed methodology is applied to the data analysis of Parkinson’s disease.Compared with non-adaptive method,random forest algorithm and others methods,the proposed method achieved higher predicted accuracy when a simplified logistic regression model is established,which shown that the proposed method has better robustness,effectiveness and interpretability.