Yeah, was the model trained using train/test splits? otherwise, the model likely has been severely overfit.

I wonder how this performance would have compared to a simple random forest or MLP model.