Research And Application Of Robust Logistic Regression

The regression analysis is an important statistical method for analyzing quantitative data,many problems can be solved by linear or generalized linear model.The results estimated by the least squares are often badly distorted by abnormal cases,even if the data set is large.Robust regression an an effective diagnostic method for detecting abnormal cases can provide unbiased estimations even if the residuals are skewed or outliers in data sets.The ordinary linear regression is not suitable for practical problems with dummy dependent variable.However,the logistic regression is the most important model with classification dependent variable,in which has an increasingly wide range of applications in many different fields.The logistic regression parameter estimation method is maximum likelihood which is very sensitive to abnormal cases and always highly susceptible to abnormal cases.Based on the analysis of three evaluation criteria of robust estimation,we studied two robust estimation methods applied to logistic regression.Compared to logistic regression,the robust logistic regression can effectively resist the interference of outliers and have better prediction in the finacial area.