Migraine Classification Using Machine Learning and Deep Learning in Low-Resource Healthcare Settings

EasyChair Preprint 15776

Abstract

Migraine is a neurological condition that impairs quality of life, with diagnostic challenges, especially in resource-limited settings lacking specialised tools and expertise. While AI models for migraine classification have been explored in standard healthcare, limited research focuses on low-resource environments. To address this, we evaluate the efficacy of Machine Learning and Deep Learning models (SVM, KNN, DT, RF, and TabNet) for migraine classification, with a focus on computational efficiency and interpretability. Among the models, RF emerged as the best model, achieving 95.8% accuracy, precision, recall, and F1 score, while TabNet achieved slightly lower performance 91.1%, 91.8%, 91.1%, 90.7% respectively. RF demonstrated enhanced computational efficiency, with a training time of 0.9s and memory usage of 0.14 MB, compared to TabNet's 10.8s and higher memory usage. Furthermore, SHAP analysis supported RF’s interpretability, and we propose RF as a cost-effective, AI-driven diagnostic tool for migraine classification, improving access to healthcare in resource-limited regions.

Keyphrases: Artificial Intelligence, Interpretability analysis, Memory Usage, Migraine, Migraine Classification, Prediction Time, Random Forest, SHAP, SMOTE, TabNet, accuracy precision recall, classifying migraine subtypes, computational efficiency, data augmentation, deep learning, low resource healthcare settings, machine learning, model size

@booklet{EasyChair:15776,
  author    = {Anithamol Ashokan and Ikram Ur Rehman and Parisa Saadati},
  title     = {Migraine Classification Using Machine Learning and Deep Learning in Low-Resource Healthcare Settings},
  howpublished = {EasyChair Preprint 15776},
  year      = {EasyChair, 2025}}