Create a normalized or standardized version of DTW

[alejandrofdez-us]

No description provided.

[alejandrofdez-us]

In assessing the suitability of Dynamic Time Warping (DTW) for comparing the similarity of time series pairs, a critical limitation arises when the series are of vastly different scales. DTW, inherently scale-sensitive, calculates the distance between points in time series while allowing for elastic transformations in the time axis. This method proves effective when comparing time series of similar magnitudes, as it accommodates for variations in speed or timing. However, when applied to pairs of time series with significant disparities in scale – for instance, one ranging in the hundreds and another in the millions – the absolute DTW distance values can be misleading. Such scenarios lead to inflated or diminished DTW values, not necessarily indicative of actual similarity or dissimilarity in the underlying patterns of the series. Consequently, direct comparison of DTW values across pairs of time series with distinct magnitude scales is unreliable, necessitating normalization or standardization of the series to a common scale before applying DTW. This step ensures that the DTW metric reflects the intrinsic temporal alignment and pattern similarity, irrespective of the absolute value differences between the series being compared.