This work presents visualization methods for the analysis of periodic or seasonal time series data sets in the univariate and multivariate case. Humans often find large data sets difficult to parse when they are plotted over a large time period. This poses a problem not only when labelling data for Machine Learning, but also when trying to explain an algorithm's prediction based on an input: In the case of periodic time series, in order to judge the results, a human would have to visually

more »

... e several data points at the same position in a period over many pattern repetitions. For this reason, we propose to split time series along period borders and remove time shifts and trend differences while enabling filtering by on-click highlighting of line segments to make outliers visible at first glance. We demonstrate our method using examples from three different domains. This method is optimized towards periodic data sets and is robust with respect to changes in trend and minor irregularities in periodicity. The method can trivially be implemented in most visualization frameworks, and is adaptable towards the specific use case and domain of new problem statements.