Machine Learning Based Cardiovascular Disease Detection Using Least Absolute Shrinkage and Selection Operator Method

Abstract

Early detection of cardiovascular disease is paramount as it stands among the most fatal and devastating illnesses globally. Despite extensive research efforts, crucial evaluation parameters like the area under the receiver operating characteristic curve (AUC-ROC), pivotal for diagnostic model assessment, have often been overlooked.  AUC-ROC, accuracy, sensitivity, specificity, precision, and other assessment metrics have also been considered in a few research, however their conclusions have not been successful. This paper introduces a model utilizing the Least Absolute Shrinkage and Selection Operator (LASSO) feature selection technique within artificial neural networks (ANN). Alongside ANN, several machine learning classifiers including Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), and K Nearest Neighbors (KNN) are employed to compare their efficacy in identifying cardiovascular illness, using a dataset of 70,000 medical records from Kaggle. According to the experimental findings, the suggested approach using ANN obtained accuracy, sensitivity, specificity, precision, and ROC score of 90.39%, 90.92%, 89.97%, 87.79%, and 0.95, respectively. In contrast, KNN, SVM, RF, and DT achieved cardiovascular diagnosis accuracies of 76.47%, 83.40%, 84.98%, and 79.58%, respectively. The findings of the classifiers utilized in this study clearly show that ANN performed better at identifying cardiovascular illness.