Business Process Monitoring with Anomaly Detection in Practice
DevOps teams within Swisscom are responsible for ensuring that any issues that could affect users are quickly addressed. This quest is challenging due to the high volume of available metrics, events, and traces (~300GB daily).
We will show how we designed and deployed an anomaly detection system based on Prophet models, PySpark, and MLflow, which analyses performance, throughput, and error metrics of important business processes. These analytics are displayed in an end-to-end monitoring tool which provides a holistic view of many distributed systems and gives actionable insights to the teams.
Joana Soares Machado
Joana is a data scientist at Swisscom. As part of the Data, Analytics and AI department, she develops ML solutions to automate the real-time monitoring of Swisscom’s business processes and IT services. Her experience includes a project at CERN where she developed a tool to monitor the responsiveness of the complex event processing engine of the ATLAS experiment. Joana holds a Master’s degree in Communication Systems from EPFL, where she previously worked as a research scientist in the domain of applied machine learning in privacy and security.
Jelena is a data scientist currently employed in Swisscom’s Data, Analytics and AI department. She finished her Computer Science master studies at EPFL and joined Swisscom firstly as an intern, and then as a full-time employee. Jelena has always enjoyed trying to make complex problems understandable to different audiences. She achieved this by working as a teaching assistant, by developing and promoting innovative topics like the Hyperloop, and encouraging younger generations of talents through initiatives like GirlsCoding. These days, in addition to improving the anomaly detection models for business process monitoring, Jelena is trying to make data science tools useful and accessible to various stakeholders.