Cardiovascular disease is a major contributor to the worldwide mortality rate, where the enormous amount of cardiac disease data generates substantial datasets with high-dimensional features. The feature selection process reduces processing time, ameliorates prediction accuracy, and aids decision-making by identifying the most pertinent features from the dataset rather than evaluating all available features. This paper proposes a population-based optimizer called Lĕvy Elephant Herding Optimization (LEHO) to address feature selection challenges; it combines the algorithms of Elephant Herding Optimization (EHO) and Lĕvy Flight (LF) to make the search space larger and more diverse, which makes it easier to find better features than with EHO alone. The LEHO optimizer relies on various machine learning models and has been applied to two datasets from Kaggle and our dataset on Coronary Artery Disease to assess its performance. Utilizing an Artificial Neural Network model, the proposed optimizer identified significant features. It attained a high average accuracy of 0.967 and 0.947 with two distinct data splitting ratios (80/20 and 70/30), respectively, for our dataset compared to the Kaggle datasets.