Detecting Outlying Time-Series with Global Alignment Kernels (Defense)
|Termin||Fr 21. Mai 2021|
|Kurzfassung||Detecting outlying time-series poses two challenges: First, labeled training data is rare, as it is costly and error-prone to obtain. Second, algorithms usually rely on distance metrics, which are not readily applicable to time-series data. To address the first challenge, one usually employs unsupervised algorithms. To address the second challenge, existing algorithms employ a feature-extraction step and apply the distance metrics to the extracted features instead. However, feature extraction requires expert knowledge, rendering this approach also costly and time-consuming.
In this thesis, we propose GAK-SVDD. We combine the well-known SVDD algorithm to detect outliers in an unsupervised fashion with Global Alignment Kernels (GAK), bypassing the feature-extraction step. We evaluate GAK-SVDD's performance on 28 standard benchmark data sets and show that it is on par with its closest competitors. Comparing GAK with a DTW-based kernel, GAK improves the median Balanced Accuracy by 4%. Additionally, we extend our method to the active learning setting and examine the combination of GAK and domain-independent attributes.