Article Preview
TopIntroduction
Heart disease is a major cause of death worldwide. About 30% of global deaths occur due to this disease. According to a survey conducted in 2015, 17.7 million people worldwide lost their lives due to this disease. According to reports from the American Heart Association, 121.5 million American adults are affected by this disease. In this disease, heart is not able to pump the required amount of blood to other body parts. Heart failure is the leading cause of human death. The symptoms of heart disease include dizziness, deep sweating and chest pain. Timely diagnosis of this disease can reduce mortality (Kumar and Rani, 2020).
Heart disease can be diagnosed using invasive and non-invasive methods. Electrocardiogram, phonocardiogram, dynamic electrocardiogram and echocardiography are some of the non-invasive methods. Non-invasive methods are less expensive and not having any side effects. But, these methods are not completely accurate (Malakar et al., 2019). So, invasive methods are usually prescribed by doctors. Coronary angiography is an invasive and most reliable method of diagnosing the disease (Jain et al., 2019). However, invasive methods are expensive and painful. Therefore, patients do not prefer these methods in the initial stages. Therefore, a system is required that can diagnose heart disease more accurately using non-invasive methods. Recent research work has shown that machine learning algorithms are having the potential to develop such systems. This type of system can be used as a screening test during disease diagnosis which reduces the burden of doctors (Rani et al., 2021a). Selection of significant features is a crucial task that increases the accuracy of such a system. Significant features can be selected using feature selection methods. Various researchers have used different types of feature selection methods to select relevant features. Filter and wrapper are two main categories of feature selection methods. Filter methods select features using general characteristics and the selected features are classifier independent. Wrapper methods select features using a classifier and the selected features are classifier dependent.
In this paper, a new Hybrid Pearson Correlation with Backward Elimination (HPCBE) feature selection method is proposed. HPCBE is developed by combining pearson correlation and backward elimination algorithms. Pearson correlation is a filter type method and backward elimination is a wrapper type method. Hence, HPCBE has combined advantages of both types of methods.
Authors have presented a hybrid system for heart disease diagnosis (HSHDD) that can diagnose heart disease using clinical parameters in which the feature selection was performed by the proposed HPCBE feature selection method. The advantage of using the purposed HPCBE method is two-fold, the classification accuracy is increased while reducing the dimensionality of the features.
There are several reasons for the motivation behind developing HSHDD. First, there is a paucity of medical resources in developing countries. So an automated system is required that can help cardiologists during heart disease diagnosis. Secondly, aggressive methods are expensive and painful. So, such a system is needed that uses non-invasive tests to diagnose the disease accurately. Third and major motivation is patients' survival rates can be increased by timely diagnosis.
The main contributions of this research work are as follow:
- (1.)
A new feature selection method called Hybrid Pearson Correlation with Backward Elimination (HPCBE) has been proposed.
- (2.)
Relevant features for the diagnosis of heart disease have been selected using the proposed HPCBE method.
- (3.)
A hybrid system for heart disease diagnosis (HSHDD) has been developed using features selected by the proposed HPCBE method.
- (4.)
Performance of HSHDD has been evaluated and compared with existing systems.
The rest of the paper is organized as follows. Section 2 contains a description of the related work done by earlier researchers. Materials and methods used for research are introduced in section 3. Results are discussed in section 4. Conclusion and future scope are given in section 5.