This paper presents a method to predict the axial compressive strength of concrete-filled double skin steel tube (CFDST) composite columns using machine learning techniques and the Comprehensive Learning Particle Swarm Optimization (CLPSO) algorithm to determine the minimal CFDST cross-sectional size required to bear a specified load. Gaussian Process Regression (GPR) and Extreme Gradient Boosting (XGBoost) algorithms are utilized. The study focuses on both short and long columns, distinguished by slenderness ratio, with input parameters including outer and inner diameters, steel tube thicknesses, yield strengths, 28-day concrete strength, and column height, and outputting the ultimate axial compressive strength. For short columns, XGBoost outperform GPR across metrics such as R-squared (R2), Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE) using 241 datasets, whereas for long columns, with 205 datasets, GPR show superior results. Optimizing cross-sectional size using CLPSO with both algorithms yield sizes closely aligning with standard design codes.