Fuzzy Clustering and Bidirectional Long Short-Term Memory for Sleep Stages Classification

This research uses feature representation with Bidirectional Long Short Term Memory (Bi-LSTM) as a final classifier. Feature learning is performed after feature extraction and aims to get the optimal represented feature. The feature representation mechanism is required as a pre-process for Bi-LSTM because Bi-LSTM is not reliable when directly processing raw data or feature extraction results. The focus of the research is to investigate the influence of cluster number of Fuzzy Clustering on Bi-LSTM performance. Specifically, the study examined the proposed method of sleep stage classification in which the data used were polysomnogram. From the testing result, it's found that increasing the number of clusters tends to increase the performance of sleep stage classification. Experiments using nine groups at the feature representation stage have the highest performance with the value of F-measure of 72.75%.