A Valid Model Evaluation Method in Epileptic Seizure Detection

Abstract

Automatic epileptic seizure detection by scalp EEG is highly demanded and researched thoroughly with many machine learning techniques. Those techniques should be evaluated in the situation that occurs in actual use scene, but previous researches did not evaluate them well. In this paper, data is split into 3 parts and used as training data, validation data and testing data, respectively. In that setting, test results are lower than validation result by 36.6% in precision and 31.5% lower in recall as the mean results of 3 models.