Event labeling combining ensemble detectors and background knowledge

摘要: Event labeling is the process of marking events in unlabeled data. Traditionally, this done by involving one or more human experts through an expensive and time- consuming task. In article we propose event label- ing system relying on ensemble detectors back- ground knowledge. The target data are usage log a real bike sharing system. We first label then evaluate performance indi- vidual labeled set using ROC analysis static evaluation metrics absence presence background Our results show that when there no access to experts, proposed approach can be effective alternative for events. addition main proposal, conduct comparative study regarding various predictive models performance, semi-supervised unsupervised approaches, train scale, time series filtering methods, online offline models, distance functions measuring similarity.