Solving Data Overlapping Problem Using A Class-Separable Extreme Learning Machine Auto-Encoder

Data overlapping and imbalanced data are significant challenges in data classification. Extreme learning machine auto-encoding (ELM-AE) is a feature reduction method that transforms original features into a new set of features capturing essential information in the data. However, ELM-AE may not effectively solve the overlapping data problem. In this research, a new method called class-separable extreme learning machine auto-encoding (CS-ELM-AE) is proposed, to improve ELM-AE's efficacy in addressing the overlapping data problem and thereby increasing classification efficiency. CS-ELM-AE encodes points in the same class of the dataset to be closer together. Oversampling is also applied to the encoded dataset to solve the imbalanced data problem. The experiments demonstrate that CS-ELM-AE could significantly improve classification model performance and achieve higher levels of accuracy, as well as greater f1-score and G-mean values than the original ELM-AE.