Comparative Study of Machine Learning and Holt-Winters Exponential Smoothing Models for Prediction of CPI’s Seasonal Data

Inflation is one of the factors influencing price stability. Inflation affects people’s purchasing power and impacts their decisions as economic actors. Consumer Price Index (CPI) is one of the factors used by economists to measure the inflation or deflation in a country. This research focuses on comparing the prediction results of the CPI for the Education Goods and Services using the Holt’s Winters Exponential Smoothing and Machine Learning Methods, namely Long Short-Term Memory (LSTM), Extreme Learning Machine (ELM), Ridge, and Least Absolute Shrinkage and Selection Operator (LASSO). The data used is univariate data on the CPI for the Education Goods and Services in Malang City in 1996-2013, which was obtained from the publication of the Statistics Indonesia (BPS), entitled “Malang City in Figures“ which was published in 1997-2014. The results of this research show that the Ridge Method produces the smallest Mean Absolute Percentage Error (MAPE) value compared to other Machine Learning Methods and the Multiplicative Holt-Winters Exponential Smoothing Method, with MAPE= 2.10723%%. Machine Learning models that have been simulated have very good accuracy values, with MAPE values $\lt10$%. Therefore, it can be assumed that the simulated Machine Learning models can make very good predictions on time-series data with seasonal patterns.