A Comparative Analysis of Multi Medical Data Classification using different Feature Selection Techniques

Abstract

By using data mining tools for early analysis and better patient survival rates, healthcare informatics is crucial in disease prediction and classification. Problems with missing value processing and choosing the best features from medical datasets, however, continue. Here, we present a full-stack approach to feature selection and classification on multi-datasets using cutting-edge optimization algorithms (Liver, Lung, Heart, and Thyroid). Modified Monarch Butterfly Optimization (MMBO), the initial algorithm to be suggested, finds the best features and fixes missing values that occur during preprocessing. A Deep Neural Network (DNN) classifier uses these properties to sort data into healthy and unhealthy norms. In multi-dataset performance tests, the MMBO-DNN algorithm achieves better results than state-of-the-art methods with regard to both accuracy and execution time. This paper presents HBSOODNN, a second model that addresses the significance of feature selection in medical data classification. This model combines an Optimal Deep Neural Network (ODNN) for classification with Hybrid Brain Storm Optimisation (HBSO) for feature selection. We optimize computation time by tuning the Brain Storm Optimization (BSO) approach with Genetic approach (GA). Next, we use a DNN that has been fine-tuned using Particle Swarm Optimization (PSO), referred to as PSO-DNN, to classify the subset with reduced features. Superb classification results are achieved by the HBSO-ODNN model on four different medical datasets. The last method is an innovative one that uses an IWD-DNN based DNN for medical data categorization and Quantum Dragonfly Optimization (QDFO) for feature selection. While IWD fine-tunes the weight and bias settings of the DNN, the QDFO algorithm chooses the best subset of features. Extensive testing on the Indian liver patient (ILP), lung cancer (LC), heart disease (HD), and thyroid datasets has demonstrated that the IWD-DNN model achieves better accuracy.