Analysis of Environmental Factors On The Risk Of Drug Abuse Using Naïve Bayes And Decision Tree Data Mining Algorithms

Abstract

This study aims to analyse and compare the performance of the Naïve Bayes classification model and Decision Tree C5.0 in predicting environmental risk factors for drug abuse in Palopo City. The research data come from the Badan Narkotika Nasional Kota Palopo, comprising a total of 74 data points with environmental label attributes and five main attributes. The data were processed through feature selection stages using Random Forest and model validation using the Leave-One-Out Cross-Validation (LOOCV) method to obtain unbiased performance estimates. The results show that personality is the most dominant attribute in determining environmental factors for drug abuse. In the performance comparison, the Naïve Bayes algorithm proved superior with an overall accuracy of 50.00% and a Mean Balanced Accuracy of 62.33%, surpassing Decision Tree C5.0 (accuracy of 47.30%). The main implication of this finding is that the Naïve Bayes model exhibits high specificity (79.39%), making it a reliable tool for early screening to validate the safety of the social environment. Practically, these findings suggest that BNNK Palopo prioritise interventions that focus on strengthening psychosocial aspects (personality) and utilise the Naïve Bayes model for more efficient and targeted resource allocation.