Adaptation of the dynamic time warping algorithm for the problem of finding the distance between two time series with periods of low value variability
Abstract
Adaptation of the dynamic time warping algorithm for the problem of finding the distance between two time series with periods of low value variability
Incoming article date: 25.01.2025The dynamic time warping algorithm (DTW) is designed to compare two time series by measuring the distance between them. DTW is widely used in medicine, speech recognition, financial market and gaze trajectories analysis. Considering the classic version of DTW, as well as its various modifications, it was found that in the tasks of analyzing the distance between gaze trajectories, they are not able to correctly take into account the duration of its fixations on visual stimuli. The problem has not attracted much attention so far, although its solution will improve the accuracy and interpretation of the results of many experimental studies, since assessing the time of visual focus on objects is an important factor in visual analysis. Hence the need to adapt DTW for such tasks. The goal of this work is to adapt the classic DTW to the problem of finding the distance between two time series with periods of low variability of values. During the demonstration of the developed algorithm, it was proven that the effect of a given minimum threshold of fixation duration on the result is significant. The proposed adaptation of DTW will improve the quality of visual data analysis and can be applied to understanding the mechanisms of human perception and decision-making in various fields of activity, such as psychology and marketing, as well as to developing effective methods for testing interfaces.
Keywords: dynamic time warping algorithm, eye tracking, time series, gaze trajectory, gaze fixation duration