Muhammad Owais Butt

Migraines are recurring and highly painful headaches with multiple associated symptoms that severely affect millions of people around the world. This condition is considered quite serious from a neurologist’s perspective because it is highly debilitating. Effective treatment of migraines begins with its diagnosis but the subjective nature of clinical evaluations along with class imbalance in patient datasets makes this very complicated. This paper attempts to tackle these issues by developing a machine-learning framework for automated migraines classification by utilizing a Kaggle dataset of 400 samples with 23 independent attributes and 1 dependent attribute representing different types of migraines. Our framework starts with a detailed cleansing of the data, which includes filtering out all missing values. Then, through the use of SMOTE (Synthetic Minority Oversampling Technique), the issue of an imbalanced dataset is tackled. This is followed by optimized feature selection through forward selection and cross-validation with Naïve Bayes. Supervised machine-learning classifiers such as Random Forest (RF), decision tree (DT), K-nearest Neighbors (KNN), and Naïve Bayes (NB) are evaluated and voted on to predict the outcome.