Application of Machine Learning Techniques for the Prediction of Heart Disease

Owodunni, Adebisi Abraham; Jaber, Tareq; Mian, Zhibao

Abstract

As important as the heart is to humans, unfortunately, 43% of death is from heart disease [2] declared by Global Burden of Disease research. By 2030, deaths from cardiovascular disease will reach 23.6 million where heart disease takes the lead [3]. Annually, 10 million people die globally according to World Health Organization (WHO). There have been (pre)established conventional ways of detecting this disease in humans like angiography, electrocardiograms among others, which are not only expensive for the common man, but have been proven, but over 17 million individuals have lost their lives to lack of expertise, incapacitation with several side effects [4]. According to a WHO survey, only 67% of the time, doctors can accurately predict heart disease. Hence the need for noninvasive and a more efficient technique thereby leveraging on Data Science (Machine Learning - ML). This research makes use of ML techniques to classifying Heart Disease through the comparative way of their metrics to predict heart disease in individuals, ii. Investigate the most relevant features and the risk factors contributing to predicting heart disease, iii. Evaluate the performance of the developed models using appropriate metrics, iv. Provide insights and recommendations for healthcare professionals to improve early diagnosis and intervention strategies. These involve four classifiers: XGBoost, Random Forest (RF), Logistic Regression (LR), and Support Vector Machine, to classify and predict heart disease using the Framingham heart disease dataset. Different models were built after handling missing values and outliers in the dataset. Before balancing the dataset, the models built, LR and RF gave the best performance with an accuracy of 85% each. The dataset was later balanced/resampled, and important features selection was done using the XGBoost classifier, Sequential Feature Selection (SFS) and KBest methods respectively, and these improved the performance of the model. Ensemble techniques (AdaBoost and Bagging) were adopted and the AdaBoost model (RF classifier) performed as high as giving an accuracy of 93%. Hyperparameter tuning was done involving Randomized SearchCV and Grid SearchCV, but none outperformed the AdaBoost model’s performance. Lastly, the balanced dataset was split into train and test datasets (ratio of 80:20), and a model was built/trained with the train dataset and then tested with the test dataset, this gave an accuracy of 93% as that of the AdaBoost model, but a better CV_score: 0.9110, R2_score: 0.7078, AUC curve: 0.98, RSME: 0.2701, MAE: 0.0730 with Random Forest classifier.

Citation

Owodunni, A. A., Jaber, T., & Mian, Z. (2024). Application of Machine Learning Techniques for the Prediction of Heart Disease. Acta Scientific Computer Sciences, 6(3), 13-23

• Journal Article Type: Article
• Acceptance Date: Feb 22, 2024
• Online Publication Date: Feb 22, 2024
• Publication Date: Feb 22, 2024
• Deposit Date: Apr 27, 2024
• Publicly Available Date: Apr 29, 2024
• Journal: Acta Scientific Computer Sciences
• Publisher: Acta Scientific
• Peer Reviewed: Peer Reviewed
• Volume: 6
• Issue: 3
• Pages: 13-23
• Keywords: Random Forest Classifier; Logistic Regression Classifier; Sequential Feature Selection; AdaBoost and Bagging; Support Vector Machine Classifier; XGBoost classifier
• Public URL: https://hull-repository.worktribe.com/output/4634452
• Publisher URL: https://actascientific.com/ASCS/ASCS-06-0510.php