Multi-Feature Learning Model for Epilepsy Classification Supervised by a Highly Robust Heterogeneous Deep Ensemble

Abstract

In this present work, we propose a novel heterogeneous deep ensemble-based multi-feature learning
environment for epilepsy classification. The proposed model is built to deal with the present prominent issues like data imbalance, low accuracy, and most importantly the need for a reliable classification model. To accomplish this, a multi-level enhancing technique is used to address the problem of class imbalance, which included data sampling with a 95% confidence interval. A variety of sample techniques are used, including random sampling, down - sampling, and the synthetic minority oversampling technique (SMOTE). We have used algorithms like significant predictor test
(SPR), Cross-Correlation Analysis (CRA) and Principle Component Analysis (PCA) to select features after retrieving samples data. The main goal of using various feature selection methods was to keep the best features for high accuracy and minimal computation. Using Decision Tree (DT), Logistic Regression (LR), Artificial Neural Network (ANN) with Radial Basis Function (RBF) and Levenberg Marquardt (LM) learning methods and Probabilistic Neural Network (PNN) algorithms as the base classifier, we created a first-of-its-kind heterogeneous deep ensemble model. Maximum Voting Ensemble (MVE) and Best Trained Ensemble (BTE) were used as ensemble decisions for two-class
classification, determining if each sample in the dataset is epileptic or not. The proposed system's superiority over main existing techniques was validated by simulation-based performance comparisons in terms of accuracy (93.88%), F-measure (0.91), and AUC (0.94).