A Deep Learning Approach for Japanese CPI Prediction: A Multi-dimensional Time Series Analysis
Year: 2025
Type: conference-paper
Abstract: This study constructs a multidimensional time series analysis model for forecasting the Japanese consumer price index (CPI) based on a deep learning approach. By analyzing historical data from 1970-2022, the study designs a forecasting framework based on LSTM and attention mechanism, which captures both time dependence and inter-category correlation of CPI. The experimental results show that compared with the traditional method, the forecasting error of this model is reduced by 32.5%, and the short-term forecasting accuracy reaches 94.2%. Energy price changes were found to have a 2-3 month leading indication for other categories. The model maintains more than 90% prediction accuracy after adding ±2% random noise, showing good stability. The research results are of great significance for improving the methodology of economic forecasting and guiding practical decision-making.
Authors: Kaifei Li, Wenjun Guo, Zhirui Pang, Yinuo He, Ping Hao, Zheyu Yang
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W4412822290