Forecasting Some Sustainable Development Goals Using Cats Swarm Optimization [CSO]: An Applied Statistical Approach on Iraq [2005–2023]

Abstract

     Iraq faces persistent unemployment fluctuating between 9.8% and 15.3% during 2005-2023, coupled with high poverty incidence (25%) and maternal mortality rates. Standard forecasting methods struggle with Iraq's economic data due to multicollinearity, non-normal residuals, and coefficient instability across conflict and oil shock periods. This study develops a hybrid forecasting model integrating Multiple Linear Regression with Cat Swarm Optimization (CSO) metaheuristic to predict unemployment from socio-economic indicators.Annual time-series data spanning 2005-2023 were analyzed using three approaches: conventional OLS regression, Random Forest machine learning, and the proposed Regression-CSO hybrid. Model performance was evaluated via RMSE, MAE, and adjusted R-squared metrics, with robustness validated through 100 Monte Carlo simulation trials.The hybrid model achieved superior predictive accuracy (RMSE = 0.41, MAE = 0.36, R²adj = 0.90), outperforming the best OLS specification by 34% and Random Forest by 29%. Poverty emerged as the strongest unemployment predictor (β = 0.573, p < 0.001), followed by GDP per capita. Monte Carlo simulations confirmed exceptional stability (SI = 0.949), demonstrating robustness across varying data conditions. Baseline forecasts project unemployment declining from 11.8% in 2024 to 10.3% in 2030, contingent on sustained poverty reduction (2% annually) and GDP growth (3.5% annually).This study demonstrates that metaheuristic-enhanced regression models offer substantial advantages over conventional methods for development indicator forecasting in contexts characterized by multicollinearity and limited data availability. The methodology is transferable to other developing countries and SDG targets with appropriate localization.