Search for: All records

Quantifying the similarity or distance between time series, processes, signals, and trajectories is a task-specific problem and remains a challenge for many applications. The simplest measure, meaning the Euclidean distance, is often dismissed because of its sensitivity to noise and the curse of dimensionality. Therefore, elastic mappings (such as DTW, LCSS, ED) are often utilized instead. However, these measures are not metric functions, and more importantly, they must deal with the challenges intrinsic to point-to-point mappings, such as pathological alignment. In this paper, we adopt an object-similarity measure, namely Multiscale Intersection over Union (MIoU), for measuring the distance/similarity between time series. We call the new measure TS-MIoU. Unlike the most popular time series similarity measures, TS-MIoU does not rely on a point-to-point mapping, and therefore, circumvents all respective challenges. We show that TS-MIoU is indeed a metric function, especially that it holds the triangle inequality axiom, and therefore can take advantage of indexing algorithms without a lower bounding. We further show that its sensitivity to noise is adjustable, which makes it a strong alternative to the Euclidean distance while not suffering from the curse of dimensionality. Our proof-of-concept experiments on over 100 UCR datasets show that TS-MIoU can fill the gap between the unforgiving strictness of the ℓp-norm measures, and the mapping challenges of elastic measures. 

Solar flares not only pose risks to outer space technologies and astronauts’ well being, but also cause disruptions on earth to our high-tech, interconnected infrastructure our lives highly depend on. While a number of machine-learning methods have been proposed to improve flare prediction, none of them, to the best of our knowledge, have investigated the impact of outliers on the reliability and robustness of those models’ performance. In this study, we investigate the impact of outliers in a multivariate time series benchmark dataset, namely SWAN-SF, on flare prediction models, and test our hypothesis. That is, there exist outliers in SWAN-SF, removal of which enhances the performance of the prediction models on unseen datasets. We employ Isolation Forest to detect the outliers among the weaker flare instances. Several experiments are carried out using a large range of contamination rates which determine the percentage of present outliers. We assess the quality of each dataset in terms of its actual contamination using TimeSeriesSVC. In our best findings, we achieve a 279% increase in True Skill Statistic and 68% increase in Heidke Skill Score. The results show that overall a significant improvement can be achieved for flare prediction if outliers are detected and removed properly.