SMOTE Approach to Imbalanced Dataset in Logistic Regression Analysis

摘要

Logistic regression is a classification model that is commonly used in bankruptcy studies. The classifier works well when data is balanced. However, imbalanced data set is found in almost all bankruptcy studies. The most common approach to deal with imbalanced data set is by selecting and matching the samples from both bankrupt and non-bankrupt samples. The problem of imbalanced data and the approach taken to deal with it can affect a good predictive model. The objective of the study is to improve the classification accuracy of a logit model when data is heavily loaded to one side. The approach taken is by using SMOTE sampling. The study used SMEs categorized under the accommodation and food service activities, and the hotel sector. There are 14 explanatory variables involved. The result from this study confirmed that the AUC and sensitivity values from SMOTE Logistic Regression (SLR) model is higher than the AUC and sensitivity values of a logit model.