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

Traditional spectral-based methods such as PCA are popular for anomaly detection in a variety of problems and domains. However, if data includes tensor (multiway) structure (e.g. space-time-measurements), some meaningful anomalies may remain invisible with these methods. Although tensor-based anomaly detection (TAD) has been applied within a variety of disciplines over the last twenty years, it is not yet recognized as a formal category in anomaly detection. This survey aims to highlight the potential of tensor-based techniques as a novel approach for detection and identification of abnormalities and failures. We survey the interdisciplinary works in which TAD is reported and characterize the learning strategies, methods and applications; extract the important open issues in TAD and provide the corresponding existing solutions according to the state-of-the-art.

A traffic tensor or simply origin  destination  time is a new data model for conventional origin/destination (O/D) matrices. Tensor models are traffic data analysis techniques which use this new data model to improve performance. Tensors outperform other models because both temporal and spatial fluctuations of traffic patterns are simultaneously taken into account, obtaining results that follow a more natural pattern. Three major types of fluctuations can occur in traffic tensors: mutations to the overall traffic flows, alterations to the network topology and chaotic behaviors. How can we detect events in a system that is faced with all types of fluctuations during its life cycle? Our initial studies reveal that the current design of tensor models face some difficulties in dealing with such a realistic scenario. We propose a new hybrid tensor model called HTM that enhances the detection ability of tensor models by using a parallel tracking technique on the traffic's topology. However, tensor decomposition techniques such as Tucker, a key step for tensor models, require a complicated parameter that not only is difficult to choose but also affects the model's quality. We address this problem examining a recent technique called adjustable core size Tucker decomposition (ACS-Tucker). Experiments on simulated and real-world data sets from different domains versus several techniques indicate that the proposed model is effective and robust, therefore it constitutes a viable alternative for analysis of the traffic tensors.

No abstract