Pendekatan Model Machine Learning dalam Pemeringkatan Status Sosial Ekonomi Rumah Tangga di Indonesia

The method used for ranking the socioeconomic status of households in the Integrated Database is to predict the value of household expenditures using the Proxy Mean Testing (PMT) method. In general, this method is a predictive model using a regression technique. The choice of statistical model used is forward-stepwise. In practice it is assumed that the predictor variables used in PMT have a linear correlation with the expenditure variable. This study tries to apply a machine learning approach as an alternative prediction method other than the forward-stepwise model. The model is built using several machine learning algorithms such as Multivariate Adaptive Regression Splines (MARS), K-Nearest Neighbors, Decision Tree, and Bagging. The results show that the machine learning model produces an average inclusion error (IE) value that is lower than the average exclusion error (EE) value. Machine learning model works effectively in reducing IE but is not sensitive enough to reduce EE. The average value of IE machine learning model is 0.21 while the average value of IE PMT model is 0.29.