Prediction of Coronary Artery Disease using Machine Learning – A Comparative study of Algorithms

Abstract

Purpose: Heart illness is one of the major killers of humans worldwide. Heart illness and the possibility of experiencing a heart attack have both increased in recent years. Medical professionals face significant difficulties when attempting to forecast heart disease. One of the medical field's virtuosi is early prediction, and this is particularly true in cardiology. The early prediction model-building studies illuminated the most up-to-date methods for locating variations in medical imaging. The study of computer-assisted diagnosis is a dynamic and quickly developing field. Since wrong medical diagnoses can lead to dangerous treatments, a lot of work has been done recently to enhance computer programs that help doctors make diagnoses. Computer-assisted diagnosis relies heavily on machine learning. The basic aspect of pattern recognition is the capability to learn from precedents. Pattern identification and artificial intelligence have a lot of promise to improve the accuracy with which biomedical professionals perceive and diagnose illness. They also help make decisions more objectively. Machine learning is a promising method for developing elegant and automatic algorithms for the study of high-dimensional and multimodal bio-medical data. Two heart disease-related datasets were considered for the purpose of this research. The study implements several machine learning algorithms and compares their prediction accuracy and a handful of other performance metrics to determine which one is the most effective. Objective: The primary goal of the research is to evaluate the performance of several machine learning algorithms using different evaluation criteria such as f1 score, roc, and auc values. The aim is to discover the most effective machine learning algorithm for the datasets obtained for the study. Design/Methodology/Approach: The research utilizes datasets from Kaggle heart information. Python, Skilearn, Pandas, and Jupyter Notebook have been used to build various machine learning prediction models and the outcomes have been compared. Findings/Results: Both datasets comprise of different parameters, therefore pre-processing had to be customized. Applying machine learning algorithms to the training dataset and comparing the trained models to the testing dataset yielded varied results for each dataset. Model performance was measured by accuracy and AUC. Both datasets gave good results with boosting algorithms, however the Cleveland dataset did better with decision trees. Originality/Value: The research included an examination of two Kaggle heart databases. It has been seen how data is distributed, how various features depend on each other, and how all the features influence the target feature of heart disease prediction. Models have been constructed and trained using different machine learning methods, each with its own set of hyper-tuning parameters. To learn which machine learning model is most effective for a given collection of data, the study has looked into both the prediction results using the trained models and the performance parameters of the individual models. Through this study, we now know more about how different machine learning methods work. To determine the most effective algorithm, it is necessary to conduct additional research of the datasets using Deep Learning techniques. Paper Type: Comparative Study