Impact of oversampling algorithms in the classification of Guillain-Barré syndrome main subtypes

Main Article Content

Manuel Torres-Vásquez https://orcid.org/0000-0001-8475-0914
José Hernández-Torruco https://orcid.org/0000-0003-3146-9349
Betania Hernández-Ocaña https://orcid.org/0000-0001-5700-7615
Oscar Chávez-Bosquez https://orcid.org/0000-0002-0324-9886

Abstract

Guillain-Barré Syndrome (GBS) is a neurological disorder where the body’s immune system attacks the peripheral nervous system. This disease evolves rapidly and is the most frequent cause of paralysis of the body. There are four variants of GBS: Acute Inflammatory Demyelinating Polyneuropathy, Acute Motor Axonal Neuropathy, Acute Sensory Axial Neuropathy, and Miller-Fisher Syndrome. Identifying the GBS subtype that the patient has is decisive because the treatment is different for each subtype. The objective of this study was to determine which oversampling algorithm improves classifier performance. In addition, to determine whether balancing the data improves the performance of the predictive models. Three oversampling methods (ROS, SMOTE, and ADASYN) were applied to the minority class. Three classifiers (C4.5, SVM and JRip) were used. The performance of the models was obtained using the ROC curve. Results show that balancing the dataset improves the performance of the predictive models. The SMOTE Algorithm was the best balancing method, in combination with the classifier JRip for OVO and the classifier C4.5 for OVA.