What is the threshold for classifying different scales of driver fatigue in classifier learner app

I'm currently working on an EEG based driver fatigue prediction system.The data sample consists of 1-hour data(1048576 samples) which is segmented into 15 minutes each(262144 samples) as non-fatigue, mild fatigue, moderate fatigue, and severe fatigue.The classification is done using classifier learner app in Matlab R2015a.Supervised classifiers such as SVM, KNN, DT, and ensemble is applied.Can anyone please tell me the threshold criteria for differentiating these 4 levels of fatigue in the classifier app

Answer 1

[Walter Roberson]

The classification learner app will not use thresholds between the fatigue levels in that system. Some human went through all of the signals and classified sections by hand, possibly based on information that is not available in the data you are provided, such as video monitoring or discussion with the driver. The data has been labeled somehow and now it is the task of the learner app to find something different in the signals between the states.

The combinations that are noticed for each of the states do not necessarily have anything odd about the individual signals. For example it might be the case that the P wave is in a certain range if you are wide awake and happy, but that if the "happy" is absent then that same P value might indicate high fatigue. Thresholds do not necessarily apply

[Ceethal Kottakali Piyus]

This helped thanks :).The classification of 15 minutes segments have done based on the oral response from the driver during the drive, is that enough for setting the threshold?

[Walter Roberson]

No. When you have multiple features then it is particular combinations that are important, not a threshold.

However, you could make the hypothesis that the determination could be reduced down to a threshold of a single signal. You would then test that hypothesis by training on individual features only and seeing how well the classification works.