Highly Correlated Linear Discriminant Analysis for Dimensionality Reduction and Classification in Healthcare Datasets

Abstract

Particularly in medical science, artificial intelligence and its applications are expanding. Accessing significant clinical data is possible, and most remain untapped. This technology will help diagnose human diseases earlier if used effectively. An effective classification system allows doctors to make a more accurate diagnosis at an earlier stage of the disease. Because medical data typically contains many features, their inclusion in decision-making processes can lead to overfitting the classification model, affecting the classification accuracy. For this reason, it is crucial to create a dimensionality reduction technique that can effectively cut down on the number of structures while simultaneously improving the classification's accuracy. To reduce the dimensionality of medical data, this article proposed Highly Correlated Linear Discriminant Analysis (HCLDA) framework. The datasets are collected from different chronic disease datasets, and missing values are analyzed using improved decision tree algorithm. The best features are selected by using weighted binary bat algorithm. The dataset clustering has used Gaussian kernel combining fuzzy c-means (GKFCM) with Particle Swarm Optimization (PSO) algorithm. The classification and dimensionality reduction (DR) have performed with HCLDA algorithm. The simulation results are compared with conventional dimensionality reduction methods, including LDA, PCA and correlation with RF classification. The classification performance metrics are compared with accuracy, sensitivity, and specificity.