ADTK is an open-source Python package for unsupervised/rule-based time series anomaly detection. It provides a unified API for common building blocks of anomaly detection models, several auxiliary tools, and includes easy access to scikit-learn models. In this talk, we will present how a data scientist may use ADTK to build a practical anomaly detection model on time-series IoT data.
ADTK is an open-source Python toolkit for unsupervised/rule-based time series anomaly detection. Focusing on building practical models in IoT environments, the toolkit offers a set of common model components with a unified API, which includes detection algorithms (detectors), feature engineering methods (transformers), and ensemble methods (aggregators), as well as pipe classes that connect them together. It also provides several auxiliary modules for data processing, visualization, model evaluation, etc.
In this talk, we will first discuss some simple, but important, concepts of time series anomaly detection that are often handled incorrectly in practice. Those discussions will lead to the motivation for creating ADTK. We then will dive into the key components of ADTK and demonstrate how to design and assemble them into a model. It will be followed by an introduction to some auxiliary modules that data scientists may find useful. Various examples will be presented in the demo of the toolkit.
An outline of the talk is as follows.