Leveraging Temporal Feature Expansion for Enhanced Prediction of Naive Bayes and Random Forest Classification on SWSR

Based on data from the Central Statistics Agency in the first semester of 2023, Central Java is one of the provinces in Indonesia with a percentage of poor people exceeding the national average rate. From these data, it can be understood that Central Java needs more attention to reduce poverty, including through effective data management of the Social Welfare Service Recipients (SWSR) database so that it can be the basis for developing social welfare service programs. Therefore, this research uses Naïve Bayes and Random Forest algorithms and combines them with a temporal feature expansion method that allows machine learning models to capture time-based patterns in the data so that the model can predict the classification of SWSR distribution in all districts/cities in Central Java for the next few years. The use of the time-based feature expansion method in machine learning classification has the advantage of identifying factors that affect future classification predictions, in contrast to time series or LSTM methods that only produce predictions without revealing these factors. The results show that the performance between the two methods is similar by getting an accuracy score of 85.71% on the best t-k model. Meanwhile, in terms of prediction length, Naïve Bayes Time-Based can predict up to the next 10 years and better than Random Forest Time-Based which is able to predict for the next 9 years. This research is expected to be able to obtain accurate and reliable prediction results to support decision-making in social welfare policies in Central Java Province.