Forecasting Government Final Consumption Expenditure in Algeria

Abstract

This study aims to compare the performance of forecasting models for final government consumption expenditure in Algeria up to 2034, using the ARIMA model based on the Box-Jenkins methodology as a traditional approach and artificial neural network models (MLP and LSTM) as modern forecasting methods. The study results showed that although the ARIMA(0,2,1) model was the most appropriate according to statistical criteria within the Box-Jenkins methodology, the artificial neural network models significantly outperformed it in forecasting accuracy, as reflected by lower error metrics (RMSE, MSE, MAE), making them effective tools for supporting strategic planning and financial policymaking. Furthermore, when comparing the two neural network models, the MLP model clearly outperformed the LSTM model, recording the lowest error metrics and the highest coefficient of determination (R²), establishing it as the optimal and most efficient model for accurate forecasting of this series up to 2034