A Combination of EDA, Machine Learning, and Artificial Neural Networks for Accurate Prediction of Heart Disease

Abstract

Cardiovascular diseases cause millions of deaths each year, with cases continuing to rise, making early prediction increasingly important. Although data science and Artificial Intelligence (AI) have been utilized to address this issue, further studies that enhance predictability and generalization are crucial, as they significantly reduce mortality rates and healthcare costs. This study employs Exploratory Data Analysis (EDA), a variety of conventional Machine Learning (ML) algorithms, and an Artificial Neural Network (ANN) to predict heart disease accurately and fill research gaps. A dataset from Kaggle, containing 1025 training samples and 303 test samples, with 14 attributes, including 13 predictive variables and a binary target indicating heart disease presence, was used. Normalization, feature importance analysis, K-fold cross-validation, and grid search were meticulously applied to improve model performance, generalization, and robustness. These methodologies led to impressive results, with most models achieving 100% accuracy, precision, recall, and F1-score on the test data, without signs of overfitting, data leakage, or bias. Principal Component Analysis (PCA) was also conducted to evaluate the richness of the features and their potential for dimension reduction. Lastly, in-depth discussions were made to clarify the study’s outcomes, compare results with the most related studies, and comprehensively examine real-world applicability.