Unsupervised Anomaly Detection of Class Imbalanced Cognition Data Using an Iterative Cleaning Method
Robert K. L. Kennedy, Zahra Salekshahrezaee, Taghi M. Khoshgoftaar
Abstract
The presence of class imbalance in machine learning datasets is a pervasive challenge that often hampers the effectiveness of traditional machine learning models. In the context of anomaly detection, the instances in the minority class are the ones of most interest. To address this issue, we evaluate an unsupervised approach that uses an iterative cleaning process for anomaly detection on cognition data. We conduct experiments on two cognition datasets, one has a large degree of class imbalance and the other is balanced. Our findings show that the unsupervised iterative cleaning approach outperforms two other unsupervised models, namely Isolation Forest and Copula-Based Outlier Detector, in the class-imbalanced dataset. The approach does not outperform both the other two models on the balanced dataset, making the approach presented particularly well-suited when there is a large class imbalance in cognition data.