Machine Learning Methods for Detecting Rare Events in Temporal Data
Nikou Günnemann-Gholizadeh · 2018
In this thesis, we address the anomaly and event detection challenge in temporal data by developing and extending intelligent data driven algorithms. By focusing on different temporal data domains, this thesis deals with detecting two categories of anomalies: (i) Local anomalies, where the goal is to find rare time intervals whose values deviate from the remaining measured data points. (ii) Global anomalies, where whole instances are detected as anomalies which statistically behave differently than the rest of the data.