Heart Disease Prediction Using Machine Learning

Background

This paper aims at analyzing the various machine learning approaches for heart disease prediction. A few relevant works in this area are discussed in this section. Heart disease is a term that assigns to a large number of medical conditions related to the heart. These medical conditions describe the abnormal health conditions that directly influence the heart and all its parts. Heart disease is a major health problem. A recent statistics about heart diseases are reported in (Go, 2014).Some preventive mechanisms for heart diseases are discussed in (Shepherd, et al., 1995). A neural network-based heart disease prediction system is discussed in (Singh, Singh, & Pandi-Jain, 2018) and the authors used 15 medical parameters of the patient. A multilayer perceptron neural network with back propagation is used to develop a trained model. A KNN based approach for remote patient monitoring is discussed in (Enriko, Suryanegara, & Gunawan, 2018). A firefly algorithm based heart prediction technique is discussed in (Long, Meesad, & Unger, 2015).

In this manuscript, we used a set of machine learning approaches for heart disease prediction. Four well-known machine learning approaches are used in this study and those are discussed in this section.

• 1.

  Naive Bayes Classifier: Naive Bayes classifier is based on Bayes theorem. It makes strong independent assumptions between features, which means every pair of features is independent compared to every other feature (Rish, 2019). In this study, the Gaussian Naive Bayes classifier is used. If the training dataset contains null values for some specific parameter then it is better to use naive Bayes classifier. The major advantages of naive Bayes classifier are listed below:

  • •

    Simple and fast

  • •

    Possible to make probabilistic prediction

  • •

    Strong independent attribute assumptions

• 2.

  KNN Classifier: The KNN stands for K-nearest neighbor and it can be used for both in classification and regression problems (Goldberger & Hinton, 2005). It is widely disposable in real-life scenarios since it is non-parametric, which means it does not make any underlying assumptions about the distribution of data. A given test data is classified by the majority vote of its neighbors. We can choose the value of K where K is the number of nearest neighbors of an object for which we wish to vote.

• 3.

  Decision Tree Classifier: The decision tree is in the form of a tree structure. It is a simple and widely used classification technique. It breaks the data into subparts, makes decisions, and at the same time decision tree is incrementally developed (Du & Zhan, 2002).