An Explainable AI approach towards Epileptic Seizure Detection

Epilepsy is a neurological condition resulting in abnormal behavior and recurring seizures. Epilepsy diagnosis requires analysis of large amount of EEG data by a trained specialist. Many potential methods for epilepsy diagnosis using EEG signals utilizing machine learning have been proposed in the literature, but understanding why a model predicts a certain output becomes important when applied to tasks in medical domain. Although the term "accuracy" may sound promising while applying on datasets, justifying a predicted output is important when it comes to diagnosis in real world. In response, we present an explainable AI approach for epilepsy diagnosis which explains the output features of a model using SHAP (Shapley Explanations) - a unified framework developed from game theory. The explanations generated from Shapley values prove efficient for feature explanation for a model’s output in case of epilepsy diagnosis. Explanations show the impact of each feature driving the output towards particular prediction. Moreover, the paper shows experimentation on different machine learning models and their SHAP explanations on the University of Bonn Dataset. The feature explanation approach demystifies the black-box algorithms and can be used to interpret a model’s prediction even in- case of false diagnosis.