I had some experiments about calculating Mahalanobis distance between each two observations by using the original data, Z-score normalised data and Min-Max normalised data, the vectors are generated from subtitles and each observation is an episode, I take the top 10 most closest distance for each observation and calculated the percentage of how many of them are from the same series (one TV series can have multiple episodes) and I got the same percentage, I can understand Mahalanobis distance takes account the covariance of the the data and Z-score normalised it, but why the result by using Min-Max normalised data are also the same? Did I miss something or Mahalanobis distance is not using Z-score? Many thanks.