Dursun, M.Güneş, S.Özşen, S.Yosunkaya, S.2020-03-262020-03-262012Dursun, M., Güneş, S., Yosunkaya, S., (2012). Comparison of Artificial Immune Clustering with Fuzzy C-Means Clustering in the Sleep Stage Classification Problem. Inısta 2012 - International Symposium on Innovations in Intelligent Systems and Applications, 2012, 1-4.9.78147E+12https://dx.doi.org/10.1109/INISTA.2012.6246976https://hdl.handle.net/20.500.12395/28747International Symposium on INnovations in Intelligent SysTems and Applications, INISTA 2012 -- 2 July 2012 through 4 July 2012 -- Trabzon -- 92831Automatic Sleep Staging is an active field of research in sleep staging community. Many methods have been applied to get rid of the cumbersome of manual staging process. Even though very effective results were taken in some of these methods with respect to the classification accuracy, they are restricted either with their limited classification data or with lower number of classified stages like wake, sleepy and deep sleep. The accuracies obtained with methods for the classification of whole sleep stages are very low to apply in real sleep staging. One reason for this is the class imbalance in training data. Approximately half of one-night sleep consists of Non-REM2 stage while Wake, Non-REM1 and Non-REM3 stages are comparatively short duration. So, the used systems can converge to the characteristics of Non-REM2 stage. Taking equal amounts of data from each stage in training can be a solution for this but in this time a question arises: which samples should be picked from the each stage. Clustering schemes can play their roles for this question. In this study, we realized this clustering process with two methods: Fuzzy C-means Clustering (FCM) and Artificial Immune Clustering (AIC). We used 55 features that extracted from the sleep EEG, EOG and EMG signals of 8 subjects. We took a total of 300 data from each stage using FCM and AIC and classified these data with Artificial Neural Networks. The performances of the used clustering methods were compared on different number of features for which PCA was applied. The results showed that AIC was over-performed to FCM by obtaining a classification accuracy of 80.62% while this accuracy was 72.16% with FCM method used. © 2012 IEEE.en10.1109/INISTA.2012.6246976info:eu-repo/semantics/closedAccessArtificial Neural NetworksC-Means ClusteringEEGSleep StageConference ObjectN/A