Identifikasi Indikator Penyebab Kemiskinan di Indonesia dengan Algoritma Decision Tree C4.5

Abstract

Poverty remains a significant issue for countries worldwide. To this day, nearly every nation, including Indonesia, grapples with this persistent challenge. In addressing poverty, it is essential for a country to develop a system capable of identifying the indicators that cause poverty, thereby enabling targeted poverty alleviation based on available resources. Consequently, this research aims to develop a system that can predict the poverty status of each Regency/City in Indonesia based on indicators hypothesized to influence poverty. Poverty status is classified into two categories: Poor and Not Poor. The dataset comprises 1,028 records, which are split into 514 training data and 514 testing data. The data undergoes preprocessing stages, including data splitting, handling missing values, data transformation, data binning, and encoding categorical variables, to ensure optimal preparation for model processing. The algorithm used is Decision Tree C4.5. The training data results indicate that the Decision Tree C4.5 model achieves an accuracy of 92.02%. Based on the Permutation Matrix evaluation to identify the indicators contributing to poverty, it is found that GDP per Capita, Expenditure per Capita, the Number of Villages with Credit Institutions, Open Unemployment Rate (TPT), and Average Years of Schooling are the most significant variables in this model.