PROJECT TITLE :

FastDTW is approximate and Generally Slower than the Algorithm it Approximates

ABSTRACT:

The repeated application of distance measure is one solution that can be used for many issues that arise when mining time series data. Similarity search, Clustering, classification, Anomaly Detection, and segmentation are some examples of the kinds of jobs that fall into this category. It has been known for more than twenty years that the Dynamic Time Warping (DTW) distance measure is the most appropriate one to apply for the majority of tasks across the majority of domains. There have been many suggestions made due to the fact that the traditional DTW algorithm has a time complexity that is quadratic. These suggestions aim to either reduce its amortized time or quickly approximate it. FastDTW is frequently mentioned as one of the most effective approximate methods. The FastDTW algorithm has been referenced in well over a thousand different publications and is directly relevant to the work of several hundred different researchers. In this work, we make a contention that may come as a surprise. The approximate FastDTW is significantly more time-consuming to use in any practical Data Mining application than the DTW itself. This fact has obvious repercussions for the user base of this algorithm, as it enables the algorithm to work with significantly larger datasets, obtain accurate results, and complete its tasks in significantly less time.