<![CDATA[Migraine is a prevalent, debilitating neurological disorder where accurate subtype classification is critical. Machine learning (ML) offers a promising avenue to enhance diagnostic accuracy. This study evaluates a Support Vector Machine (SVM) model for multi-class migraine classification. Utilizing a public Kaggle dataset, data was partitioned into 75% training and 25% testing sets. An SVM with a linear kernel was implemented to classify seven migraine subtypes. Performance was evaluated using overall accuracy, a confusion matrix, and detailed per-class metrics: Precision, Recall, and F1-Score. The model achieved 82.65% overall accuracy and a weighted-average F1-Score of 0.824. However, detailed metrics revealed significant variance. The model achieved perfect F1-Scores (1.000) for 'Migraine Without Aura' and 'Typical Aura without Migraine' but struggled with class confusion. 'Typical Aura With Migraine' exhibited a low Recall (0.533), and 'Basilar-Type Aura' had a poor F1-Score (0.400). Critically, the model completely failed to classify 'Sporadic Hemiplegic Migraine' (0.000 F1-Score), a failure masked by the high overall accuracy. These results suggest the linear SVM is a viable baseline, but its reliability varies drastically across subtypes. The granular F1-Score and Recall metrics are essential, exposing classification failures hidden by overall accuracy. Future work must address class imbalance and symptomatic overlap, likely via non-linear models, before this approach is clinically viable.]]>
