Publication

Abstract : Cardiovascular diseases are the prime cause of global deaths. The World Health Organization statistics point cardiovascular diseases as having nearly half of the non-communicable diseases. Many researchers have designed and proposed various computerized prediction models for detecting heart diseases early. A deep neural network model is proposed in this paper, using four hidden layers to detect coronary heart diseases. Three different combinations of input layer—hidden layer—output layer were evaluated and the best model is proposed. The proposed model focuses on avoiding overfitting. The datasets used are Statlog and Cleveland datasets available in the UCI Data Repository. Accuracy, sensitivity, specificity, F1 score, and misclassification are the different metrics used to evaluate the proposed model. Further, ROC is plotted with AUC. The model gave promising figures of 98.77% (accuracy), 97.22% (sensitivity), 100.00% (specificity), 98.59% (F1 score), and 1.23% (misclassification) for the Statlog dataset. For the Cleveland dataset, the corresponding values are 96.70%, 92.86%, 100.00%, 96.30%, and 3.30%, respectively.