Identifying the Predictive Capability of Machine Learning Classifiers for Designing Heart Disease Detection System

The diagnosis of heart diseases through invoice based techniques as well as ordinary medical based methods are not reliable. On other hand, non-invoice based techniques are more effective for heart disease diagnosis. Therefore, we check the capability of various Machine Learning (ML) classifiers and deep learning classifier for heart disease identification in this paper. Six machine-learning classifiers and BPNN were used in order to check which one classifier is more effective for diagnosis the heart disease. The feature selection algorithm Relief was used for selection of important features and on these selected features, classifiers performances were also computed. Ensemble machine learning techniques (boosting, bagging, stacking) were used to further increase the classifiers performance. Furthermore, cross-validation techniques k-folds was also used. Additionally backward propagation neural network (BPNN) was also used for classification purpose because deep learning algorithm not need feature selection algorithms and it automatically select important features for good result. Based on model performance evaluation metrics the SVM (RBF) performed excellently on full features achieved accuracy 86%, and 88% accuracy on selected features as compared other classifiers. Through Ensemble learning techniques, SVM obtained the classification accuracy 92.30%. The BPNN achieved 93% classification accuracy. Thus the performance of deep neural networks learning algorithm is better than traditional machine learning algorithms. As per our experimental results shows that the performance of BPNN based diagnosis system is more effective for heart disease diagnosis