Deep Learning-Based Time Series Forecasting: A Convolutional Neural Network Approach for Predicting Inflation Trends

Abstract

This study investigates the application of Convolutional Neural Networks (CNNs) for forecasting inflation trends in Egypt, aiming to enhance the accuracy of economic predictions by capturing complex, non-linear temporal dependencies in time series data. Traditional econometric models, such as ARIMA and VAR, often struggle to model volatile and dynamic economic conditions, prompting the exploration of deep learning techniques. The proposed CNN-based model leverages historical inflation data from 1960 to 2023, sourced from the World Bank, to predict future inflation trends. The methodology involves data preprocessing, feature extraction using convolutional layers, and prediction through fully connected layers, optimized using the Adam optimizer. Performance metrics, including Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and the Coefficient of Determination (R²), demonstrate the model's robustness, with an RMSE of 9.2113 and an R² of 0.8911 on the testing dataset. The results indicate a steady upward trend in inflation from 2024 to 2030, with rates rising from 12.45% to 16.35%, accompanied by widening confidence intervals reflecting increased uncertainty over longer horizons. The study concludes that CNNs offer a reliable framework for inflation forecasting, outperforming traditional methods in capturing non-linear patterns. Recommendations include integrating additional economic indicators and exploring hybrid models to further enhance predictive accuracy. This research contributes to the growing application of deep learning in economic forecasting, providing valuable insights for policymakers and researchers.