RNN and PSO Based Evaluation of Weights of Cardiac Affecting Risk Factors

This paper presents, weight optimization of cardiac affecting parameters (such as blood pressure, body mass index, diabetes, blood oxygen saturation level, stress level, life style etc.) using an efficient combination of recurrent neural network (RNN) algorithm and particle swarm optimization (PSO) technique. Since cardiovascular diseases lead to significant global health challenge and still the leading cause of death. Therefore, this study is useful to detect early signs of heart disease by analyzing cardiac affecting parameters. This work includes the opinions of 50 doctors regarding 16 cardiac affecting parameters. Each doctor assigns weights to cardiac-affecting parameters, based on their expertise in cardiology. The variability in doctors’ opinions on assigning weights to each cardiac affecting parameter creates confusion. The inconsistency in doctors’ opinions on various cardiac affecting risk factors (CARFs) makes it difficult to determine the ideal cardiac weight values. To address this cardiac challenge, a hybrid (RNN+PSO) technique is employed for cardiac-affecting parameters’ weight optimization. The approach involves selecting the global best (gbest) for each cardiac parameter. Considering these gbest values as the optimized value, and subsequently converting it from optimized values to normalized values through a weight normalization formula. The hybrid (RNN+PSO) model achieved an average accuracy of 85.21% and an average loss of 14.79% during weight optimization for CARFs. The most important normalized optimized weights for cardiac-affecting parameters are evaluated as: very high blood cholesterol level (0.093), body mass index (0.074 for obese class), high heart rate (0.71), and high blood cholesterol level (0.070). This hybrid model will help to cater the prior diagnosis of patients with cardiovascular diseases, indicating risk factors, and possible treatment options.