Ischemic heart disease remains a major contributor to mortality in Malaysia, with non-elective percutaneous coronary intervention (PCI) frequently performed in high-risk acute coronary syndrome (ACS) patients. Using nationwide registry data (2007-2020), we evaluated 29,521 patients and compared seven machine learning (ML) models for predicting in-hospital, 30-day, and 1-year mortality. Models were developed in a training cohort and externally validated using hospital-level (TEST1) and prospective temporal (TEST2) cohorts. After logistic recalibration, discrimination for in-hospital mortality ranged from 0.927 to 0.943 (TEST1) and 0.865-0.884 (TEST2). For 30-day mortality, ROC-AUC ranged from 0.902 to 0.923 (TEST1) and 0.753-0.838 (TEST2), and for 1-year mortality from 0.833 to 0.859 (TEST1) and 0.750-0.801 (TEST2). Calibration remained acceptable, and decision curve analysis demonstrated positive net benefit across clinically relevant thresholds. Cross-model stability analysis consistently identified age, haemodynamic status, and renal function as key predictors.