| dc.description.abstract | This study applies various machine learning models to forecast Taiwan’s Consumer Price Index (CPI), Core Consumer Price Index (Core CPI), and Producer Price Index (PPI), comparing the performance of different models over various forecast horizons. The models used include Autoregressive (AR), LASSO regression, Elastic Net regression (EN), Random Forest (RF), Deep Neural Network (DNN), Recurrent Neural Network (RNN), and hybrid models (LASSO+Elastic Net and Elastic Net+Random Forest).
The results show that the AR model performs excellently across all forecast horizons, especially in long-term forecasts, demonstrating good stability and accuracy. It is the best choice for both short-term and long-term forecasts. The RF model also performs well in short-term and medium-term forecasts, showing strong predictive capabilities, particularly in handling high-dimensional data and capturing complex patterns in the data. The LASSO and EN models perform poorly in short-term forecasts but improve in long-term forecasts. DNN and RNN exhibit some potential in long-term forecasts but have higher errors in short-term forecasts. Hybrid models are less stable than individual models across different forecast horizons, but they show some improvement in long-term forecasts. | en_US |