Forecasting the Consumer Price Index in the Regions of the Philippines using Machine Learning for Time Series Models

Abstract

The core objective of this study is to showcase the enhanced forecasting capabilities of a hybrid model that combines the strengths of Artificial Neural Networks (ANN) and Autoregressive Integrated Moving Average (ARIMA) in predicting the Consumer Price Index (CPI). By harnessing the intricate non-linear pattern capturing ability of ANN and the capabilities of ARIMA in modeling linear and autoregressive components, the hybrid model aims to outperform the standalone ARIMA model in accurately forecasting the CPI. Real-world CPI data will be utilized for empirical evaluation and comparison, providing valuable insights into the effectiveness and practical applicability of the hybrid ARIMA-ANN approach in improving CPI forecasting accuracy. The performance of Box Jenkins Models which gives the resulted value of R-squared values for both stationary and non-stationary data are high, indicating that the models explain a significant portion of the variability in the CPI data. The RMSE, MAPE, and MAE values are relatively low, suggesting that the Box-Jenkins models' predictions are close to the actual values. The Ljung-Box Q statistic indicates that all Box-Jenkins models best fit their respective CPI data. The study also employs rigorous statistical methods of machine learning model accuracy assessment, including the Akaike Information Criterion (AIC), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE), to assess the forecasting performance of both models. The results demonstrate that the hybrid ARIMA-ANN model consistently outperforms the standalone ARIMA model, delivering more accurate and reliable forecasts over an extended forecast horizon. The integration of Artificial Neural Networks (ANN) using Multilayer Perceptron (MLP) in the ARIMA models improved the accuracy of the fitted and forecasted values. RMSE and MSE values for the Hybrid ARIMA-ANN models are lower compared to the original Box-Jenkins/ARIMA models, validating the goal of enhancing accuracy through ANN integration.