Machine Learning Application for Classification Prediction of Household’s Welfare Status

Abstract

Various approaches have been attempted by the Government of Indonesia to eradicate poverty throughout the country, one of which is equitable distribution of social assistance for target households according to their classification of social welfare status. This research aims to re-evaluate the prior evaluation of five well-known machine learning techniques; Naïve Bayes, Random Forest, Support Vector Machines, K-Nearest Neighbor, and C4.5 Algorithm; on how well they predict the classifications of social welfare statuses. Afterwards, the best-performing one is implemented into an executable machine learning application that may predict the user’s social welfare status. Other objectives are to analyze the reliability of the chosen algorithm in predicting new data set, and generate a simple classification-prediction application. This research uses Python Programming Language, Scikit-Learn Library, Jupyter Notebook, and PyInstaller to perform all the methodology processes. The results shows that Random Forest Algorithm is the best machine learning technique for predicting household’s social welfare status with classification accuracy of 74.20% and the resulted application based on it could correctly predict 60.00% of user’s social welfare status out of 40 entries.