An Innovative Model Proposal for Forecasting Budget Revenues in Turkiye: XGBoost-LSTM-GRU Hybrid Approach

Abstract

This study compares the budget revenue forecasting performance of various machine learning and deep learning models, such as Multilayer Perceptron (MLP), Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM), Extreme Gradient Boosting (XGBoost), and a stacked XGBoost-LSTM-GRU hybrid model, and comprehensively forecasts future monthly budget revenues from November 2023 to December 2024 with the model with the best forecasting performance. In order to evaluate the training and testing performance of the models, budget report data published by the Ministry of Treasury and Finance for the period January 2008-October 2023 is used. The study demonstrates the potential of sophisticated analytical models in financial management and draws important implications for improving fiscal planning and policymaking.