Heart Disease Prediction Using Machine Learning Approach

Abstract

Heart diseases have become the primary cause of death globally. Therefore, it is essential to develop robust diagnostic and treatment methods. This thesis focuses on diagnosing heart disorders. We utilized the MIT-BIH Arrhythmia Dataset to conduct a comparative analysis of various machine learning (ML) techniques, including Random Forest (RF), K-Nearest Neighbour (KNN), and Decision Tree (DT), along with deep learning (DL) models such as Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM). To enhance predictive performance, various preprocessing methods were employed, including filtering, normalization, and comprehensive feature selection techniques like chi-square and sequential feature selector. Additionally, an advanced prediction was proposed, combining feature selection using a hybrid of Genetic Algorithm (GA) and Cuckoo Search Optimization (CSO) with a majority voting ensemble of Convolutional Neural Network and Random Forest on UCI Heart disease dataset. This approach also integrated GA for hyperparameter tuning, enhancing predictive accuracy. Comprehensive preprocessing techniques were employed to ensure data quality, including handling missing values, outlier detection, and normalization. The results demonstrate that our method outperforms traditional models. This study contributes to advancing predictive analytics in cardiovascular healthcare, aiming to support early diagnosis and informed decision-making processes through robust and accurate predictive models.