Prediction of heart disease is a major issue to begin and personalize the treatments and maximize the benefit of those treatments. Heart disease is a major noticeable illness that increases heart disease by influencing several risk factors faced by the people such as age, high blood pressure, cholesterol, blood sugar, etc. The risk of diabetes increases constantly with the increasing nature of glucose levels and it causes heart disease. The conventional prediction model is a great deal for predicting heart disease with the cause of diabetes at an earlier stage. But achieving a higher degree of accuracy rates using conventional algorithms and feature selection methods is a difficult task. In order to improve the accuracy of heart disease digenesis, a novel deep learning technique called Hot Deck Imputed Robust Congruence Convolutive DEep Neural Learning (DICDEN) technique is introduced. The proposed DIRCOL technique consists of three different processes namely data-preprocessing, feature selection, and classification. The number of patient data related to diabetes is collected from the dataset. Then the data preprocessing is carried out using Manhattan hot deck imputed technique to transform raw data into the structured format by handling the missing value and data duplication. After the data preprocessing, the Rand indexive robust linear regression is employed in the proposed DICDEN technique for significant feature selection and removing the other irrelevant features. Followed by, classification of data samples is done with the selected features by using a deep learning technique called Tucker Congruence Radial Damped Convolutive Deep Learning Classifier. The proposed classifier includes numerous layers such as one input layer, multiple hidden layers, and one output layer. First, the number of selected features with the training patient data is given to the input layer. Then the input is transferred into the hidden layer where the feature mapping is performed in the convolution layer using the Tucker congruence correlation coefficient. The radial activation function is applied to provide the final disease classification results at the output layer. In order to minimize the error, the damped least square method is applied. Finally, accurate classification results with a minimum error are obtained at the output layer. Based on the classification results, heart disease is correctly predicted. Experimental evaluation is carried out with different quantitative metrics such as disease prediction accuracy, precision, recall, F-measure, and disease prediction time. The analyzed results reveal the performance of our proposed DICDEN technique when compared with the existing deep learning methods.

Heart disease predicting, Manhattan hot deck imputed technique based data preprocessing, Rand indexive robust linear regression based feature selection, Tucker Congruence Radial Damped Convolutive Deep Learning Classifier