Analysis of Cardiovascular Disease Prediction Using Model-Agnostic Explainable Artificial Intelligence Techniques

Introduction

Cardiovascular diseases (CVDs) also known as heart diseases are a leading cause of death worldwide, with an estimated 17.9 million people dying from CVDs. Cardiovascular diseases are a range of heart and vascular problems (Westerlund et al., 2021), (Virani et al., 2021). According to a World Health Organization (WHO) report, more than four out of five cardiovascular disease deaths are caused by myocardial infarction, and one-third of these deaths occur in people under the age of 70. Identifying high-risk individuals for cardiovascular disease and ensuring appropriate treatment can help prevent early deaths. The majority of CVDs can be prevented by avoiding risk factors such as excessive alcohol consumption, a poor diet, obesity, cigarette use, and physical inactivity (Ghosh et al., 2021).

Figure 1.

Graph displaying relationship between age and heart disease frequency

Figure 1 illustrates which age group are suffering more from heart diseases. The blue color of the graph indicates the people (in millions) not suffering from heart disease and orange indicates the people suffering from heart disease. It can be depicted that in the age group 29-54, people suffering from heart diseases are more compared to not suffering from the heart diseases. In contrast, for the people whose age is greater than 54, the people suffering from heart diseases are comparatively less.

Figure 2.

Graph displaying relationship between gender and cardiovascular disease percentage

Both the Figures 1 and 2 show general trends of occurrence of heart diseases with respect to age and gender.

Numerous machine-learning algorithms, such as the Support Vector Machine, the Decision Tree, and numerous more, have been suggested. While these algorithms provide great accuracy, they are not easily interpretable. In comparison, although techniques such as Naive-Bayes and Linear Regression can be studied, their accuracy is limited (London, 2019), (Duval, 2019). While they are capable of analyzing sophisticated algorithms, they lack the ability to relate the data's properties (Kavila et al., 2021), (Pawar et al., 2020). Given that the majority of individuals are unable of comprehending sophisticated algorithms, it is critical to create a model that is understandable to the average person.

In healthcare, models must explain why they made a particular categorization and significantly reduce their accuracy numbers to do so. These are referred to as Explainable AI (XAI) approaches (Porto et al., 2021). We used the SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) to implement these methods. The Model-Agnostic XAI technique's characteristics are intended to help visualize the qualities' contribution to classification. In this study we used methods such as Support Vector Machine with Normal Distribution Model (SVM-NDM), Logistic Regression (LR), K-Nearest Neighbors and Single-Layered Artificial Neural Networks (SL-ANN) and Multi-Layered Artificial Neural Networks (ML-ANN) to create an end-to-end interpretable Explainable Artificial Intelligence system for cardiovascular disease prediction. Our Major Contributions outlines:

• 1.

  Predicting the cardiovascular disease of the data points using machine learning and neural network models.

• 2.

  Displaying the interpretability behind the prediction with the help of SHAP and LIME technique plots.

The outline of the paper is as follows: Section 2 represent the overview of related research on Cardiovascular disease prediction. Subsequently, section 3 sketches out the proposed system and Section 4 describes the results and accuracies of the models and plotting model-agnostic explainable artificial intelligence techniques plot for a data point. Conclusion and Recommendations for future works are in section 5.