• The FA–IPSO–SVM model achieves high prediction accuracy for converter endpoint carbon content, with hit rates up to 98.74% within ±0.02% error.
• Factor analysis effectively reduces data dimensionality while preserving predictive performance, outperforming several existing methods.
• Input parameters are systematically classified into high, medium, and low influence categories based on their impact on prediction accuracy, aiding feature selection.
• The study addresses a gap by analyzing the influence of input parameters, not just improving model algorithms, providing practical guidance for industrial applications.