Practical Guide To Anomaly Detection
Anomaly detection is a mystery: you can recognize it, when you see it with your eyes, but you can’t very strictly formalize it mathematically. In other words – you don’t know what you are looking for exactly, since you don’t know the characteristics of anomalies. This is an unsupervised-learning approach of anomaly detection. Rarely, when you know what is anomaly (supervised case) – then it becomes a classification problem with unbalanced classes.
Anomaly detection has a wide range of useful applications: manufacturing (reliability), IoT, cyber security, fraud detection.
In my talk I will focus more on models for unsupervised spatial data. I will give their description and comments when it is more suitable to use them. Methodologies for supervised and temporal data will be also covered.
Senior Data Scientist and Solution Architect, GE Healthcare
Valentin Mikhaylenko is Senior Data Scientist and Solution Architect is GE Healthcare. He has worked on different analytical projects: Text Mining, Predictive Maintenance, Labor Optimization, Chatbot.
Prior to that, Valentin was software developer in automated stock trading company (High Frequency Trading) for 7 years building the whole trading platform from scratch and Data Scientist in Natural Language Processing startup for 2 years.
He graduated summa cum laude from ITMO University (Saint Petersburg, Russia) with a MSc in Applied Mathematics and Informatics. Also Valentin graduated from Borland Academy (AMSE, supported by Yandex and Jetbrains).