Comparison of Error Rate Prediction Methods in Binary Logistic Regression Model for Balanced Data

Abstract

Binary Logistic Regression is one of the statistical methods that can be  used to see the relations between dependent variable with some independent variables, where the dependent variable split into two categories, namely the category declaring a successful event and the category declaring a failed event. The performance of binary logistic regression can be seen from the accurary of the model. Accuracy can be measured by predicting the error rate. One method that can be used to predict error rate is cross validation. The cross validation method works by dividing the data into two parts, namely testing data and training data. Cross validation has several learning methods that are commonly used, namely Leave One Out (LOO), Hold out, and K-fold cross validation. LOO has unbiased estimation of accuracy but take a long time, hold out can avoid overfitting and works faster because no iterations, and k-fold cross validation has smaller error rate prediction. Meanwhile, data cases with different correlation are useful to find out the different correlations effect performance of error rate prediction method. In this study uses artificially generated data with a normal distribution, including univariate, bivariate, and multivariate datasets with various combination of mean differences and correlation. Considering these factors, this study focuses on comparing the three cross validation methods for predicting error rate prediction in binary logistic regression. This study finds out that k-fold cross validation method is the most suitable method to predict errors in binary logistic regression modeling for balanced data.