Perbandingan Naive Bayes dan Support Vector Machine dalam Klasifikasi Tingkat Kemiskinan di Indonesia

Abstract

Poverty in Indonesia is a complex issue influenced by various economic and socio-cultural factors. This study aims to compare the performance of Naïve Bayes and Support Vector Machine (SVM) in classifying poverty levels in Indonesia while also evaluating the effectiveness of random oversampling in addressing data imbalance. The dataset consists of 514 samples from various districts and cities in Indonesia, with 452 samples classified as "not poor" and 62 as "poor." After applying oversampling, the total number of samples increased to 730, with a balanced distribution (365 samples per class). The observed data include socio-economic indicators such as the percentage of the poor population, per capita expenditure, the Human Development Index, and the open unemployment rate. The study splits the data using an 80:20 ratio for training and testing. Experimental results show that SVM achieved a higher accuracy of 81% compared to naïve bayes, which reached 76%. Additionally, SVM demonstrated a more stable balance between precision and recall. On the other hand, the oversampling technique effectively improved the model’s ability to identify the minority class, particularly for Naïve Bayes, which was more responsive to data duplication. These findings highlight the role of machine learning in designing more effective social policies for poverty data management.