Machine Learning based Cardiovascular Disease Prediction Using One-vs-one Approach

Among the various chronic diseases that pose a threat to human life, cardiovascular disease (CVD) is the most rampant one. The rapid growth in mortality rate due to CVD has made it a matter of concern for the healthcare practitioners all over the world. Besides finding suitable forms of treatment, another important challenge faced by them is handling the massive amounts of patients data generated in the healthcare centres in various forms through Electronic Health Records (EHRs). Solutions to both these challenges are provided by the use of machine learning models. Machine learning models help in identifying different useful patterns hidden in the enormous datasets. Also, they help in predicting the chances of occurrence of CVD in a person well-in-advance with the help of certain health parameters, thereby eliminating any perilous condition. Most of the machine learning models developed so far classify the patients into two classes: disease present or disease absent. Comparatively lesser models have experimented on their multiclass classification. It is important to detect the level of impact of CVD on a person along with the detection of its presence or absence. In this research work, we have studied upon the Cleveland HD dataset to classify patients into five classes ranging from 0 to 4. Further, the Logistic Regression (LR), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) algorithms are implemented in their original forms and the performances are compared by implementing a One-vs-One (OVO) approach in terms of the recall provided. It was observed that the highest recall of 77.2% was provided by the OVO implementation of KNN algorithm.