Comparative Study of Extreme Gradient Boosting (XGBOOST), K-Nearest Neighbors (KNN), and Random Forest for Migraine Classification

Migraine is a common neurological disorder that can seriously compromise the quality of life of the affected individuals. Migraine's typical diagnosis is solely dependent on traditional diagnostic methods which relies on patient self-reporting and clinical judgment, which can be subjective and prone to errors. The main objective of this study was to model migraine classification using Extreme Gradient Boosting (XGBoost), Random Forest, and K-Nearest Neighbors (KNN) algorithms, integrating Least Absolute Shrinkage and Selection Operator (LASSO) for feature regularization. Through this study, the classifications abilities of these machine learning models were evaluated to determine which among them is superior in terms of classifying the type of migraine one is suffering from. To prevent overfitting and enhance interpretability, LASSO regression was utilized for feature regularization. The models were trained with a labeled data set, hyperparameter tuning was achieved through Grid Search to systematically explore different combinations of hyperparameters and identify the optimal settings that maximize models performance. The models were evaluated based on accuracy, precision, recall, ROC-AUC, F1-score and computation time. The top-performing model was deployed into a web-based application using Spring Boot. XGBoost outperformed the other models, achieving an accuracy of 92.4%, an AUC of 96.0%, an F1-score of 91.65%, and a sensitivity of 92.24%, with a false positive rate of 1.59% and a computation time of 2.08s. Random Forest followed closely with 91.6% accuracy, a 94.0% AUC, an F1-score of 90.49%, and a sensitivity of 86.45%, but required 4.65s of computation time. K-Nearest Neighbors (KNN) demonstrated the lowest performance, with an accuracy of 86.6%, an AUC of 91.0%, F1-score of 80.53%, a sensitivity of 79.32%, and the highest computation time of 9.51s. XGBoost was found to be the most appropriate choice for migraine classification. This study highlights the promise of machine learning in enhancing migraine diagnosis through objective and data-driven means.