Refining the census income prediction through the development of random forest regressor and assessing performance with gradient boosting regression

Income prediction using census data is a crucial aspect of socio-economic research, aiding in policy-making and resource allocation. This study aims to refine income prediction models by developing a Random Forest Regressor (RFR) and evaluating its effectiveness against Gradient Boosting Regression (GBR). This research introduces a comparison between ensemble learning techniques through which it evaluates their performance when analysing complex socio-economic datasets with high dimensions. The main goal is to improve predictive model robustness while enhancing accuracy through implementation of feature selection and hyperparameter adjustment with cross-validation processes. The database contains demographic along with socio-economic data that goes through a comprehensive preprocessing phase for imputing missing values while converting categories to numerical forms and normalizing them. The RFR model benefits from grid search and k-fold cross-validation to reach optimization while its performance evaluation compared to GBR model occurs through Mean Absolute Error (MAE), Mean Squared Error (MSE), and R-squared (R 2 ) values. The results indicate RFR achieves superior performance over the models compared to GBR because it manages nonlinear patterns and controls overfitting effectively. Research establishes RFR emerge as the most favourable approach for income modelling in intricate datasets along with GBR which demonstrates strong capability. The research outcomes advance machine learning applications for socio-economic modelling which serves as fundamental knowledge for data-driven policies development.