Search

Anomalies in time series, also called "discord," are the abnormal subsequences. The occurrence of anomalies in time series may indicate that some faults or disease will occur soon. Therefore, development of novel computational approaches for anomaly detection (discord search) in time series is of great significance for state monitoring and early warning of real-time system. Previous studies show that many algorithms were successfully developed and were used for anomaly classification, e.g., health monitoring, traffic detection, and intrusion detection. However, the anomaly detection of time series was not well studied. In this paper, we proposed a long short-term memory- (LSTM-) based anomaly detection method (LSTMAD) for discord search from univariate time series data. LSTMAD learns the structural features from normal (nonanomalous) training data and then performs anomaly detection via a statistical strategy based on the prediction error for observed data. In our experimental evaluation using public ECG datasets and real-world datasets, LSTMAD detects anomalies more accurately than other existing approaches in comparison. [ABSTRACT FROM AUTHOR]

• Effectively detect anomalies in multivariate time series (MTS) is challenging. • In practical applications, few algorithms are widely used. • A space-embedding strategy for anomaly detection (SES-AD) in MTS was developed. • SES-AD model works efficiently in benchmark datasets and real datasets. • SES-AD outperformed other classical approaches. Anomaly detection of time series has always been a hot topic in academia and industry. However, many existing multivariant time series methods suffer from common challenges due to increased dimensionality. In this study, we developed a space-embedding strategy for anomaly detection in multivariant time series (SES-AD). As a hybrid model, SES-AD did not directly search the discords from the original time series, but projected the raw sequence to a lower dimensional space so that the significant abrupt change points in the new space can be easily captured from the dissimilarity vector. Finally, the potential abnormities were determined by a statistical strategy. To verify the performance of our method, SES-AD was applied to an extensive number of multivariate time series. The experimental results suggest that SES-AD is more efficient than five existing approaches. Overall, the SES-AD model is suitable for solving anomaly detection for high-dimensional datasets and guarantees computational effectiveness and accuracy. [ABSTRACT FROM AUTHOR]