Hybrid feature selection and Optimized Deep CNN for Heart disease Prediction

The main cause of death in the world is heart disease. Accurate detection of heart illness is critical for competently managing cardiac patients prior to a cardiac arrest. Moreover, the volume of information composes manual prediction and analysis taxing and time-consuming. The early diagnosis of people in hazard level for the disease is essential for avoiding its growth. A Deep Learning (DL) approach is better to predict heart disease. Deep Convolutional Neural Network (Deep CNNs) is widely used for medical decision support to accurately detecting and diagnosing various diseases. Because of their capability to identify the relations and concealed designs in health care data, DCNNs have been exceedingly successful for designing health support systems. The Min-max normalization technique is developed in this stage of preprocessing. In addition, the Kumar-Hassebrook and Dice coefficients are used in the feature selection process. This method uses embedded feature selection to choose a subset of structures, which are considerably related with a heart disease. Bootstrap is a broadly applied and really powerful analytical tool for data quantification. A Light Spectrum optimization (LSO)-based technique has attained maximum values of accuracy, sensitivity, and specificity of 95 %, 94.9 %, and 93.8 % for 90% of learning set.