Comparison of ARIMA and Deep Learning Models for Forecasting the Consumer Price Index in Uzbekistan: Using R Packages
Abstract
For The Consumer Price Index (CPI) holds significant importance in macroeconomic analysis and decision-making, serving as a crucial indicator for detecting and regulating the general price level as well as contributing to national economic accounting. The study aimed to develop optimal models to predict the daily number of new cases, with the goal of assisting in the formulation of strategies related to infection. The autoregressive integrated moving average (ARIMA) models and deep learning models: neural networks (RNNs) models, long short-term models (LSTM), and gated Recurrent Unit (GRU) models—were utilized to fit the daily new cases. The performances were assessed using the minimum mean absolute percentage error (MAPE). The predictive performance of the trained model is detailed in Table 4–5, encompassing various metrics including RMSE, MAE, and MAPE. Examination of the performance measures in Table 4-5 reveals that the GRU model achieves the highest scores in terms of RMSE, MAE, and MAPE. AdaBoost, on the other hand, outperforms others in R2, a finding supported by Fig. 8. Additionally, LSTM emerges as the optimal choice for J2. These results underscore that the model with the best performance on absolute/relative metrics, such as RMSE and MAPE, may not necessarily coincide with the most efficient training method or the most stable approach with the least variance.