Table of contents- Introduction --Definition --Anomaly Types
Special anomalies
Level shift anomaly
Detection using rolling aggregate and double rolling aggregate
Pattern change anomaly
Volatility shift anomaly detector
Anomaly detection by threshold method
Three different methods
Anomaly detection GENERAL CASE & METHODS
Exploratory data analysis
Three different approaches Approach 1. Move, smoothe, evaluate 1.Rolling window estimations 2.Exponential smoothing, Holt-Winters model 3.Time-series cross validation, parameters selection
Approach 2. Econometric approach- ARIMA 1.Stationarity, unit root 2.Getting rid of non-stationarity 3.SARIMA intuition and model building
Approach 3. Linear (and not quite) models on time series 1.Feature extraction 2.Linear models, feature importance 3.Regularization, feature selection 4.XGBoost Conclusion and future work