Hybrid Optimization based Feature Selection with DenseNet Model for Heart Disease Prediction

Abstract

The prevalence of cardiovascular diseases (CVD) makes it one of the leading reasons of death worldwide. Reduced mortality rates may result from early detection of CVDs and their potential prevention or amelioration. Machine learning models are a promising method for identifying risk variables. In order to make accurate predictions about cardiovascular illness, we would like to develop a model that makes use of transfer learning. Our proposed model relies on accurate training data, which was generated by careful Data Collecting, Data Pre-processing, and Data Transformation procedures. Additionally, the optimal selection is carried out on the existing attributes by fusing two meta-heuristic procedures, the Lion Algorithm (LA) and the Butterfly Optimization Procedure (BOA), into a single method dubbed the hybrid Lion-based BOA (L-BOA). In this research, we analyse the amount of parameters in a deep learning model and provide an end-to-end solution for classifying patients as healthy or unwell. To extract deep features from the best data, the proposed approach makes use of pre-trained convolutional neural networks-(CNNs) dubbed DenseNet121. The model can benefit from a more nuanced feature set composed of the features derived from each CNN. Accuracy, precision, recall, and the F1-Score were used to rate the trained classifiers. The models' classification results demonstrated that the inclusion of pertinent characteristics significantly improved the classification precision. When compared to models skilled on a full feature set, the performance of organization replicas trained with a smaller feature set improved dramatically with less training time.