ODIN AD: a framework supporting the life-cycle of time series anomaly detection applications Niccolò Zangrando [0000-0002-4796-5649] , Piero Fraternali [00000-0002-6945-2625] , Rocio Nahime Torres [0000-0003-2865-0278] , Marco Petri [0000-0001-5368-9196] , Nicolò Oreste Pinciroli Vago [0000-0001-7906-4987] , and Sergio Herrera [0000-0002-8903-0622] Department of Electronics, Information, and Bioengineering, Politecnico di Milano, 20133 Milan, Italy {niccolo.zangrando, piero.fraternali, rocionahime.torres, nicolooreste.pinciroli, sergioluis.herrera}@polimi.it, marco.petri@mail.polimi.it Abstract. Anomaly detection (AD) in numerical temporal data series is a prominent task in many domains, including the analysis of indus- trial equipment operation, the processing of IoT data streams, and the monitoring of appliance energy consumption. The life-cycle of an AD application with a Machine Learning (ML) approach requires data col- lection and preparation, algorithm design and selection, training, and evaluation. All these activities contain repetitive tasks which could be supported by tools. This paper describes ODIN AD, a framework assist- ing the life-cycle of AD applications in the phases of data preparation, prediction performance evaluation, and error diagnosis. Keywords: Time series · Anomaly detection · Data annotation · Model evaluation · Evaluation metrics. 1 Introduction With the advent of IoT architectures, the analysis of numerical temporal data series is being increasingly applied in such industries as manufacturing and con- struction, in which machines, appliances, and whole systems are equipped with sensors producing timestamped numerical data streams. Applications include anomaly detection (AD) [5] whose primary focus is to find anomalies in the op- eration of working equipment at early stages to alert and avoid breakdowns. AD is also at the core of predictive maintenance (PdM) [29], which aims at optimiz- ing the trade-off between run-to-failure and periodic maintenance, improving the Remaining Useful Life (RUL) of machines, and avoiding unplanned downtime. The development of an AD solution follows the typical life-cycle of a data-driven application, illustrated in Figure 1. Such a workflow differs from that of a tra- ditional software system because it relies on predefined parametric algorithms that must be fit to the specific task and data at hand [7].