1
$\begingroup$

So I trained a DL algorithm using Keras for Human Action Recognition. The model has an accuracy of about 85 percent and a loss of 0.3 something. The problem is that the model did not predict well on unseen data. The dataset consists of 3 classes with 100 videos in each class.

Training Phase: Training of DL model The model evaluation [Model evaluation][2]

This was a handwaving video that was predicted as walkingThis was a handwaving video that was predicted as walking The modelThe model

$\endgroup$