Finding relevant features for identifying subtypes of Guillain-Barré Syndrome using Quenching Simulated Annealing and Partitions Around Medoids
Keywords:clustering, search optimization
We present a novel approach to find relevant features for identifying four subtypes of Guillain-Barré Syndrome (GBS). Our method consists of a combination of Quenching Simulated Annealing (QSA) and Partitions Around Medoids (PAM), named QSA-PAM method. A 156-feature real dataset containing clinical, serological and nerve conduction test data from GBS patients was used for experiments. Different feature subsets were randomly selected from the dataset using QSA. New datasets created using these feature subsets were used as input for PAM to build four clusters, corresponding to a specific GBS subtype each. Finally, purity of clusters was measured. Sixteen features from the original dataset were encountered relevant for identifying GBS subtypes with a purity of 0.8992. This work represents the first effort to find relevant features for identifying GBS subtypes using computational techniques. The results of this work may help specialists to broaden the understanding of the differences among subtypes of GBS.