CPI Index Prediction Based on SARIMA-LSTM-CatBoost Model

This article seeks to enhance the predictive accuracy of the Consumer Price Index (CPI) by addressing the challenges faced by individual models in capturing the linear, nonlinear, and cyclical characteristics present in data sequences. We introduce a machine learning-based combined model that combines Seasonal Autoregressive Integrated Moving Average (SARIMA) and Long Short-Term Memory (LSTM) models. The raw CPI data is first transformed from year-over-year comparisons to a fixed base index. Thereafter, we train the SARIMA and LSTM models to generate individual forecast outcomes. These forecasts are refined through the application of the CatBoost algorithm, which optimizes the weighting of the predictions to produce a consolidated forecast. A comparative analysis with single models and other combined models derived from various machine learning techniques demonstrates that the SARIMA-LSTM-CatBoost model excels in predictive performance, significantly reducing the Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). This confirms the superior precision and broad applicability of the SARIMA-LSTM-CatBoost combined model in CPI forecasting. Our research not only presents a novel approach to CPI prediction but also equips economic analysis and policy-making with a powerful analytical tool.