A Novel Heart Disease Prediction System using Deep Multi-Layer Perceptron and Optimal Feature Selection Mechanism

Diagnosis and prognosis of heart disease (HD) are essential medical tasks for a correct classification, which helps cardiologists to treat the patient properly. The current medical system is unable to obtain the entire information from the heart disease database. It is difficult for a physician to analyze and diagnose chronic disease because it is a challenging endeavor. Hence this paper proposes a novel weight and bias tune deep multi-layer perceptron for heart disease prediction (WBTDMLP) with optimal feature selection using modified random forest (MRF). The proposed system comprised ‘3’ phases such as data preprocessing, feature selection, and HD prediction. Initially the HD prediction data is collected from the Cleveland dataset and the missing value imputation and data normalization is applied on the dataset to preprocess the dataset. Following that, the feature selection was performed by using the MRF algorithm. Finally, the HD prediction is done based on WBTDMLP approach and the parameters are tuned by Sobel sequence with Brownian random walk-based dragonfly optimization algorithm (SSBRWDOA). The results indicate that the proposed approach reaches 97.89% accuracy, which is relatively higher than existing methods.