ADASYN Oversampling and SGB Ensemble Algorithm for Migraine Classification

Abstract

A migraine is a neurological disorder that causes severe headaches. Although it is not life-threatening, it greatly affects people's lives. Migraine classification is considered a complicated research area. Early detection and accurate classification help in determining the appropriate care, speeding recovery from the disease, and avoiding its effects. So, developing a migraine classification system is very necessary. The aim of this work is to suggest a migraine classification model based on a stochastic gradient-boosting ensemble algorithm. First, the dataset is pre-processed. Adaptive Synthetic (ADASYN) is used to balance a migraine dataset that contains seven groups of migraine patients. Twenty-three migraine attributes were extracted from 400 patients. The ANOVA feature selection method is applied to select the most relevant features. Four experiments have been carried out based on apply/not apply ADASYN or ANOVA. The SGB model is trained using the number of hyperparameter tunings used in the four experiments in order to improve model performance. The SGB model is evaluated in terms of precision, recall, f1-score, and accuracy. The results showed that SGB achieved outstanding outcomes in the fourth experiment when applying ADASYN and ANOVA. The model achieved 97.01% accuracy, 0.970905 precision, 0.966250 recall, and a 0.967278 f1-score.