Unified Explanations in Machine Learning Models: A Perturbation Approach