Hybrid MLP-CatBoost Model for GDP Growth Forecasting: A Machine Learning-Driven Economic Decision Support System

Accurate prediction of GDP growth is technically complicated, due to the complexity of nonlinear relations and robust temporal dependencies in macroeconomic indicators. Traditional machine learning models are unable to capture long-term dependencies across different countries. To overcome these technical challenges, this research proposes a Hybrid Multilayer Perceptron MLP CatBoostRegressor model. The MLP extracts complex features from lagged features and moving averages; the whole CatBoost model captures nonlinear interaction and minimizes overfitting through gradient boosting. The strength of this integration enables the model to learn temporal dependencies and macroeconomic relationships. The model was trained on a global economic dataset released by the World Bank up to 2024, which includes features such as inflation rate, Unemployment Rate, Interest Rate, and Stock Index Value, used for efficient decision-making. The proposed hybrid framework achieved exclusive performance metrics, including an R-squared value of 0.976, a Mean Squared Error of 0.001, a Mean Absolute Error of 0.031, and a Pearson correlation coefficient of 0.988, which considerably exceeded those of baseline architectures such as MLP (R-squared = 0.8771) and DNN (R-squared = 0.923). These results ensure that the model's higher capability in capturing macroeconomic trends, minimising overfitting, and delivering reliable, high-precision GDP forecasts over diverse economic conditions.