Prediction Model for Detection of Heart Disease Stages Using Machine Learning Approaches.

Abstract

Health sector reports reveal that heart patients are exponentially growing concerning time. Although there are several types of medication and treatment mechanisms available in the health sector. However, researcher’s efforts are trying to explore the applications of machine learning as an emerging area of research. The healthcare sector has accumulated enormous amounts of data that may contain some hidden insights or useful health indicators, which may later be useful for effective decision-making in treatment and medication processes. In this research, an effective heart disease prediction system model is developed using machine learning approaches. The designed system model for the smart prediction about the risk stage (risk level) of heart disease can be a new and alternative instrument in the medical sector for heart disease diagnosis. This research used 14 attributes like Sex, Age, Chest Pain, Family History, Past History, Cholesterol, Fasting Blood Sugar, Resting ECG, Slope ST, Heart Rate, Pulse Rate, Blood Pressure, CBC, and Diagnosis. The proposed model predicts the likelihood or the appropriate stage of the risk of heart patients.  This mechanism can help doctors to minimize the later-stage risks and consequences. These hidden factors or patterns are required to be discovered and analyzed to reveal the hidden insights or patterns from the unused and unexplored data especially in Ethiopia hospitals. Initially, it was aimed to categorize the status of heart disease in terms of major or minor stages of risks.  We used the KDD process model to find out and interpret the discovered patterns from data repositories. Decision trees (J48 and Random Forest), Bayes (Naïve Bayes), and ANN (Multilayer perceptron) algorithms were used for the classification of data mining tasks. After experimentation, the overall accuracy of the tested classifiers was achieved 90% + in approximation. It was revealed that the ANN (multilayer perceptron) classifier relatively produces higher classification accuracy (97.0%) than the other selected four classifiers. The classification was based on the obtained attributes and revealed that prediction rates are not uniform among all the classifiers and the selected attributes. Finally, the study concluded that machine learning can be used as a new technique to discover the hidden patterns or insights in the heart patient’s massive amount of data to determine the risk stage and this can help in minimizing the risks at an early stage.