Predictive Model Comparison for Predicting Condom Use: Comparison of Conventional Logistic Regression and Other Machine Learning

Abstract

Condom use at first sex remains an important issue as it shapes future sexual behavior. This study aimed to deploy and predict condom use using five different machine learning classification models. Dataset used for this study was from Indonesian Demographic and Health Survey (IDHS) 2017 with a population of interest was male adolescents. We evaluated five different models, namely logistic regression, naïve bayes, K-Nearest Neighbors, support vector machines, and decision tree. Performances of each model were assessed using metrics such as accuracy, specificity, sensitivity, ROC Curve, and AUC Score. Study found that different models exhibit different accuracy, specificity, sensitivity, ROC Curve, and AUC Score. The decision tree and naïve bayes models remained the models with the highest specificity and sensitivity, however the KNN model expressed the highest AUC score. Result from the conventional logistic regression also explained that condom use was associated with education level, age at first sex, and attitude towards condom use. The government is advised to create equal education opportunities for every adolescent and shape better knowledge and condom attitudes. Future studies are advised to enhance the performance of machine learning models using hyperparameter tuning and other methods.