A metaheuristic-enabled training system for ensemble classification technique for heart disease prediction

Abstract

Various conditions that affect the muscles, blood arteries, heart, valves, or internal electrical pathways that regulate muscle contraction are referred to as "heart diseases." This work propose a new heart disease prediction model. The three main steps of the suggested framework are feature extraction, proposed dimensionality reduction, and proposed optimal ensemble-based heart disease prediction. Here, the most crucial components of the collected input data were extracted first. For accurate prediction, a significant number of features have been retrieved. Using Normalized Mutual Information induced Principal Component Analysis (NM-PCA), a novel feature dimension reduction technique has been proposed to overcome these problems. The dimensionally reduced features obtained by NM-PCA are simultaneously trained by the SVM, RF, and KNN classifiers. The improved optimized RNN generates the final prediction result. This optimized RNN is trained using the relevant outputs from SVM, RF, and KNN. Deer Updated Moth flame Optimization is used to adjust the weight function of the optimized RNN since it makes the ultimate judgment (DUMFO). After utilizing DUMFO, the RNN's prediction accuracy has greatly increased. The Deer Hunting Optimization Algorithm (DHO) and the Moth Flame Optimization (MFO) algorithms are combined to generate the DUMFO. Finally, using both positive and negative criteria, the performance of the suggested methodology is compared to existing, methodologies. The accuracy of the projected model is 47.3%, 45.2%, 57.8%, 36.8%, 26.3%, 14.7%, 16.8%, 5.12% and 3.15% superior to existing models like DBN, NN, CNN, LSTM, PM-LU, GA, SMO+EC, SSO+EC, BOA+EC, WOA+EC, MFO+EC and DHOA+EC, respectively.