Prediction of Heart Disease Using Machine Learning with Data Mining

Introduction

Machine learning offers strategies, tactics, and resources that can assist in resolving diagnostic and prognostic issues in a range of medical specialties. The significance of clinical indicators and their combinations for prognosis is examined using machine learning. It has developed into a method that is often used to gather medical data for things like planning treatments, outcome studies, and estimating illness progression. Additionally, machine learning is employed for data analysis in the form of smart alerts, continuous data interpretation in intensive care units, and the replication of inaccurate or missing data based on pattern discovery in the available data. It is well recognized that when machine learning techniques are successfully deployed, they aid in the integration of computer-based healthcare systems, present chances to facilitate and enhance the work of medical professionals, and ultimately improve the effectiveness and caliber of medical care [1]. Machine learning is being used in the healthcare industry not to replace doctors, but to reduce their workload and give patients feedback more quickly and efficiently.

Machine learning has several important applications in the field of medical diagnosis [2]. In this approach, doctor- based techniques are used to create hypotheses from patient data. In order to do this, the system is enhanced with symbolic learning techniques and knowledge management capabilities that are appropriate for the doctor’s interpretation of the case. As a result, the forecast is generated using straightforward rules or, most often, a decision tree. The reporting of a medical imaging as a certain radiologist using machine learning is an illustration of this.

Biomedical signal processing is an additional application area [3]. With the use of machine learning techniques, it is feasible to model the linear or non-linear relationships that exist between the data and find the fundamental features and information that are concealed in physiological signals or that are likely to be disregarded. Additionally, machine learning is employed in radiography, magnetic resonance imaging, endoscopy, confocal microscopy, computer tomography, and other imaging techniques, particularly for the detection of cancerous regions. In addition to all of these uses, the most typical application of machine learning is to forecast diseases using categorical classification algorithms and patient data [4, 5, 6].

In addition to increasing daily, cardiovascular disorders make up a sizable share of non- communicable diseases. More than a third of all deaths are caused by cardiovascular illnesses, which constitute the leading cause of death worldwide [7]. Coronary heart disease, cerebrovascular disease, hypertension, congenital heart disease, peripheral artery disease, rheumatic heart disease, and inflammatory heart disease are all examples of cardiovascular illness. Tobacco use, physical inactivity, a poor diet, and abusing alcohol are the main causes of cardiovascular disease [8]. Rapid diagnosis has become more crucial due to the rising disease incidence and death.

The goal of this study was to use some demographic factors and biomarkers in combination with machine learning to categorize those with no history of heart disease (healthy) and those who have been diagnosed with heart disease (patients).

Results and Discussion

The study examined age, sex, cp, trestbps, chol, fbs, restecg, thalach, exang, oldpeak, slope, ca, and thalach data to investigate the presence of heart disease using certain biomarkers. While these parameters are within the typical reference range for healthy people, they do not match the reference range for sick people. Adults who are healthy are supposed to have no cp complaints, fbs values under 120 mg/dl [15], resting heart rates between 60 and 100 [16], no exang symptoms, and chol values under 200 mg/dl [17]. While there is a significant difference in age, sex, cp, trestbps, restecg, thalach, exang, oldpeak, slope, ca, and thalach values between healthy and sick persons in the data set used in the study (p<0.05), there is no significant difference in the chol and fbs values (p>0.05).

Ten different randomly generated fitts were explored in order to identify the biomarkers that contribute significantly to the AIC value for disease prediction. The glm.fit method is employed in this step to enhance machine learning. The biomarkers and AIC data included in each foot are displayed in Table 3.

It is advised to choose the model with the lowest AIC value for predicting the disease. Due to this, the estimation was performed using the biomarkers from glm.fit.1, which produced the lowest AIC value (237.48). It was found that age data from biomarkers contributed little to the prediction.

Machine learning was carried out with MLR in the form of healthy/patient prediction using sex, cp, trestbps, chol, fbs, restecg, thalach, exang, oldpeak, slope, ca, and thal biomarkers. Table 4 contains the complexity matrix that was produced as a result of the estimation.

The classification of healthy and heart sick people using the MLR algorithm of machine learning was found to be 88% successful. The precision value, which shows how many of the values we predicted as positive are actually positive, was calculated as 93%. The sensitivity and specificity were found to be 86% and 91%, respectively. Table 5 displays a comparison of studies that classified heart disease using various methods and the same data set.

Chaurasia, et al. [7] performed heart disease classification with popular algorithms such as CART (Classification and Regression Tree), ID3 (Iterative Dichotomized 3) and decision table (DT) and reported that they achieved 83.49%, 72.93%, and 82.50% accuracy rates, respectively. By classifying using deep learning with artificial neural networks, the hybrid random forest with a linear model, and many other models, Mohan, et al. [18] recommended the adoption of the hybrid random forest with a linear model (HRFLM), with the greatest accuracy rate of 88.4. Taşçı, et al. [19] classified cardiac disease in their experiments using well-known algorithms as Naive Bayes, Random Forest, Decision Tree, Logistic Regression, and SVM. They claimed that the Naive Bayes algorithm stood out among all of these techniques with an accuracy of 88.52%.

The MLR algorithm was used in this study to classify cardiac disease, and the results were found to be 88% accurate and consistent with the literature.

Conclusion

In this study, instead of using complex or multiple algorithms as in the literature, the biomarkers used in classification were rearranged. In this way, only the markers with a high contribution to the classification were included in the study and 88% accuracy was obtained by using a single algorithm called MLR. Considering the clinical intensity, it has been seen that the MLR algorithm, which is easy to apply, provides fast and highly accurate results, will be useful in estimating heart disease.