EXPLAINABLE MACHINE LEARNING FOR EARLY MIGRAINE DETECTION AND SEVERITY CLASSIFICATION USING SMOTE-BALANCED ENSEMBLE MODELS

Abstract

Migraine is fundamentally a neurological disorder that triggers extremely severe headaches. Typically appears on one side of the head. Common symptoms include sensitivity to light and sound, nausea and vomiting. The headache can last anywhere from a few hours to a couple of days. It may impact day to day functioning of a person. The most confusing or tricky part is that its symptoms are the same as symptoms of other diseases. That's why we develop a system that can detect migraine at an early stage and even determine what kind of migraine a person has. This we determine from the symptoms. We got the data from a hospital in Soledad, which is also available on the Kaggle. The dataset includes information on about 400 patients with 23 different medical features. Data was not balanced as some types of migraine were rare. In order to correct this data, we apply a technique called SMOTE which, essentially, generates new samples to level out the data to 451. Various machine learning methods are applied, such as Random Forest, XG boost and begging classifiers. After cleaning and scaling the data, we use the Random Forest model with SMOTE to work the best. The training of this model has two aims: The first one is to find out if a patient has a migraine or not and the second one is to classify the type of migraine like mild, moderate, or severe. It provides almost 99% accuracy for both. The precision, recall and F1-score all had high scores. This helps to make the information far more reliable for doctors to use in making decisions. All things considered, this system is light, speedy and precise. Suitable for real-time medical applications.