For people living with migraine, the uncertainty of when the next attack might occur can be a burden in itself. A new study suggests that artificial intelligence could help predict the risk of a migraine attack as much as one day in advance.
The study, published in Neurology Open Access, analyzed data from more than 53,000 people with migraine and over 770,000 daily records collected between January 2020 and July 2025.
The Model Achieved 81% Accuracy
Researchers tested seven machine-learning algorithms to determine which could most accurately predict whether a person would experience a migraine the following day.
The selected model achieved an overall accuracy of 81%. Its positive predictive accuracy was 91.2%, while sensitivity reached 80% and specificity 83%.
This means the model was able to identify, with a considerable degree of accuracy, both days when a migraine episode was expected and days when one was not.
The researchers noted that the analysis included headaches of varying intensity, ranging from mild and moderate to severe, rather than focusing exclusively on the most intense attacks.
Not Just Today’s Symptoms, but the Previous 30 Days
One of the study’s most notable findings concerns how the risk of a future migraine may be predicted.
Although migraine attacks can be preceded by symptoms such as fatigue, nausea, mood changes, frequent yawning, neck stiffness, food cravings, and sensitivity to light or sound, these symptoms were not the most important source of information for the model.
Instead, the individual’s headache history over the previous 30 days carried greater predictive weight. In particular, the average intensity of headaches during the previous month was identified as the single most important factor in predicting a migraine attack.
Overall, characteristics derived from the previous 30 days of headache history accounted for approximately 56% of the model’s predictive information, while warning symptoms contributed around 11%.
