Classification Using Nonparametric Logistic Regression for Predicting Working Status

Abstract

Logistic regression is classical and prominent method for classification and it is used as benchmark for comparing the alternative methods. However, logistic regression is not always superior compared to the other methods. The accuracy of logistic regression could be improved by incorporating nonparametric model. The response variable used in this study is working status of housewife that categorized as working or not-working. Meanwhile the predictor variables consists of three variables, they are highest education level, age, and household expenditure. The result of fitting model shows that by incorporating nonparametric model to the binary logistic regression model can improve the classification accuracy. This is indicated not only by accuracy percentage, but also by area under Receiving Operating Characteristic (ROC) curve. The dataset will be divided into two parts, 80% as training data and 20% as testing data. The classification accuracy resulted by the binary logistic regression model is 60.36% for training data and 59.30% for testing data. Meanwhile, the classification accuracy of nonparametric logistic model is 63.43% for training data and 64.94%. for testing data. The classification accuracy and area under curve of nonparametric logistic regression is higher than those of binary logistic regression.