Exploiting the Relationship Between Kendall's Rank Correlation and Cosine Similarity for Attribution Protection