Toolkit for Time Series Anomaly Detection

Dhaval Patel, Dzung Phan, Markus Mueller, Amaresh Rajasekharan
2022 Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining  
Time series anomaly detection is an interesting practical problem that mostly falls into unsupervised learning segment. There has been continuous stream of work being published in top-tier data mining and machine learning conferences. We invented many anomaly algorithms [1, 2, 14] , procedures [6, 7, 10] , frameworks [4, 12, 13] and applications [3, 5, 8, 9, 11] while working on real industrial application settings. This tutorial presents a design and implementation of a scikit-compatible
more » ... for detecting anomalies from time series data for the purpose of offering a broad range of algorithms to the end user, with special focus on unsupervised/semisupervised learning. Machine Learning pipeline (MLPipe) become an interesting artifact in data driven research and industrial community. MLPipe orchestrates the series of steps that needs to be performed on incoming data while conducting a model training as well as model scoring. MLPipes are defined for various machine learning tasks such as classification, regression, imputation, clustering, outlier detection, etc. MLPipes are also defined for various data modality such as iid, time series, images, test, etc. In this tutorial, we introduce Anomaly Pipeline for time series data covering both univariate and multi-variate data. Compared to traditional MLPipe, the anomaly detection pipeline for time series data differs in various ways: (1) unsupervised model training; (2) real valued anomaly scores to label generation scheme; (3) supporting for high dimensional data (more than 100+ sensor variables); (4) multiple intent of anomaly such as point anomaly, segment anomaly, and contextual anomaly. Given an input time series, the tutorial will discuss how data scientist can construct four categories of anomaly pipelines followed by an enrichment module that helps to label anomaly. In particular, four types of anomaly pipelines will be discussed: RelationshipAD, WindowAD, DeepAD and ReconstructAD. These pipelines provide a platform to discuss the different mechanisms that exist in the literature to build the anomaly pipeline. Irrespective of pipeline categories, the structure of pipelines remains same such as transformer, estimator, etc. and we will review these commonly used components.
doi:10.1145/3534678.3542625 fatcat:zpgradijo5g7dc3sbejot35eau