Yakamoz: A Novel Time Series Anomaly Detection Method
Abstract
In today's society, data accessibility and quality are becoming increasingly important. Hence, detecting abnormal instances in data is a crucial task for many different disciplines. The primary objective of our work is to provide unsupervised, model-free, accurate and efficient detection of anomalous observations in time series data. For this reason, a novel time series anomaly detection method has been proposed. Anomalous samples are not required in the training datasets for this method. Given that collecting labelled data can be challenging, this ability is extremely practical. Additionally, the performance of the method is unaffected by the location of the anomaly. Using real and simulated time series datasets, the success of the proposed method in detecting anomaly/anomalies is investigated. The performance comparison results indicate that the proposed method is very effective in detecting anomaly/anomalies in these time series regardless of the variation in the data.