An Enhanced Extreme Learning Machine Based on a Swarm Intelligence Approach for Heart Disease Detection

Abstract

Heart disease (HD) is frequently considered the most problematic human disease. Designing a more precise heart disease prediction technique is a challenging task. Recently, several health disease detection problems have been addressed using the extreme learning machine (ELM) approach. The ELM has become increasingly generally employed in various learning domains for prediction and control applications due to its rapid learning rate, straightforward structure, and excessive generalizability. Finding the perfect input weights and hidden bias parameters is the foremost challenge to enhancing the ELM's performance. Randomly selecting these parameters wants more hidden neurons than the traditional gradient learning technique, delaying the prediction response and decreasing the likelihood of finding the ideal output weight. In this study, bacterial colony optimization (BCO) is used to adjust the connection weights and bias of ELM (BCO+ELM) to address the drawbacks above. Additionally, the missing value of the heart dataset is filled with multivariate imputation by chained equation (MICE) and relevant features are selected by using recursive feature elimination (RFE) for obtaining more efficient solution accuracy and enhancing the performance of proposed BCO+ELM. According to the performance measures, BCO+ELM shows excellent prediction performances. The experimental effects indicate that the BCO+ELM creates better outcomes and low computation time with fast convergence time when compared to other approaches. The best-performing BCO+ELM shows consistent gains, although at a somewhat smaller rate (0.4% after imputation, 2.76% after feature selection). This implies that BCO+ELM still gains from the improvements provided by both imputation and feature selection, further optimizing its performance even if it is already highly optimized.