Applying Machine Learning to Predict the Consumer Price Index in Saudi Arabia

Abstract

One of the key economic indicators is the consumer price index (CPI) which has frequently been used by financial policy makers of several countries to predict inflation (or deflation). Using guidelines from various studies, a majority of central banks have began adopting monetary policies targeted as inflation which call for a stringent requirement of an accurate prediction model of the CPI. However, the prediction accuracy achieved by numerous studies is low and must be improved. This study presents an attempt to forecast the future CPI of Saudi Arabia based on the previous years' CPI data. It develops a matrix comprising monthly CPI data derived from the General Authority for Statistics covering the period of January 2002 to December 2018 to be used as a case study and subsequently carries out simulations in WEKA, where seventy percent (70%) of the data is used to train the model and thirty percent (30%) is used for testing. Furthermore, the study uses the root mean squared error (RMSE), with average error rate of 0.436, the mean absolute error (MAE), with average error rate of 0.281, the relative absolute error (RAE), with an average error rate of 8.963, the root relative squared error (RRSE), with an average error rate of 11.653, and the correlation coefficient (CC), with an average error rate of 0.987, as error metrics for a performance evaluation. The second part of the study is conducted using simulations in Mathematica and CPI data of Saudi Arabia from 1963 to 2019 to predict yearly future CPI using the gradient boosted trees prediction algorithm, which provides sufficiently accurate predictions. The gradient boosted trees prediction is also compared with other prediction algorithms.