It would be nice if you explain why you use the meanSquaredError as loss and not a categoricalCrossEntropy or binaryCrossEntropy (since in your case it would be a binary classification). Thank you!