<p>Angina Pectoris (AP) is a serious condition or discomfort in chest when heart does not get enough blood and oxygen. It is one of the silent killers and hence timely diagnosis is needed for saving lives. Thus, the given paper introduces a cost-sensitive framework called Clinically Informed Feature-Weighted Loss XGBoost with Regularized Attention (CIFWL-XGB-RA) to forecast AP. The framework handles false negatives which are more serious concern to the patients for severe consequences. Clinical misclassification costs are incorporated along with attention mechanism, enabling the framework to focus on the most important clinical features. Furthermore, the proposed framework ensures sensitivity to high-risk cases and effectively addresses class imbalance. With the deployment of class weight calibration and attention regularization, CIFWL-XGB-RA achieves better sensitivity and interpretability together. The proposed model demonstrates superior performance over state-of-the-art methods, achieving 92.28% accuracy, 90.43% precision, 89.72% recall, and 90.97% f1-score when tested on large-scale simulations using real clinical datasets. The results demonstrate model’s performance regarding improvement in decision-making processes in patient outcomes.</p>