Enhanced hybrid classification model algorithm for medical dataset analysis

Abstract

The medical industry generates a significant volume of data that requires effective machine learning models to make accurate predictions for public healthcare. Current Machine Learning (ML) techniques have limitations in feature extraction and classifier accuracy. In this paper using diabetes dataset classification, to address these issues, propose a novel algorithm that enhances Hybrid Classification Model approach by integrating advanced methods tailored for high-dimensional medical data. To handle Missing Values (MV) and outliers, a hybrid imputation approach that combines K-Nearest Neighbor (KNN) and Multivariate Imputation by Chained Equations (MICE) is initially used to preprocess the datasets. Feature extraction (FE) is performed using Deep Feature Extraction techniques, including Convolutional Neural Networks (CNNs) and Autoencoders, followed by Feature Fusion to create a comprehensive feature set. For Feature Selection (FS), introduce an Advanced Ensemble Feature Selection method employing Genetic Algorithm-Based Feature Selection (GAFS), Multi-Objective Evolutionary Algorithm (MOEA), and Relief-Based Methods to identify the most relevant features. Finally, classification is achieved through a Hybrid Classification Model incorporating Ensemble of Classifier with Stacked Generalization (Stacking), Boosting, Bagging and Neural Network (NN) Enhancements with attention mechanisms (AM) and Transfer Learning (TL). This integrated approach enhances the robustness and accuracy of medical data classification. Comparing the suggested approach with current methods, the experimental outcomes show a considerable improvement in accuracy (A), sensitivity (S), specificity (SP), and reduced execution time (ET).