In today's world, cardiovascular diseases are recognized as one of the leading causes of global mortality. Early diagnosis of these conditions using machine learning techniques can play a vital role in reducing risk and improving treatment quality. This article examines and compares standard methods for predicting heart disease based on the UCI Heart Disease dataset, which includes 920 records and 16 features. Baseline methods such as Random Forest, without any advanced feature engineering, achieve an accuracy of around 75%. In contrast, the proposed approach, by incorporating newly engineered features such as a composite risk index, age grouping, the heart rate-to-age ratio, and BMI estimation, and by optimizing the model using GridSearchCV and an automated pipeline, achieves over 85% accuracy. These innovations not only reveal hidden patterns in the data but also reduce model uncertainty through permutation importance and cross-validation. The results show a 10% improvement in F1-score and a significant reduction in false negatives. Ultimately, it is recommended that similar innovations be applied to other heart-disease-related datasets to help develop more accurate and reliable clinical decision-support systems.