Evaluating the Robustness of ML Models to Out-of-Distribution Data Through Similarity Analysis