Dunning T., Friedman E. Practical Machine Learning. A New Look at Anomaly Detection

New York: O’Reilly, 2014. — 66 p.Looking Toward the Future
The Shape of Anomaly Detection
Finding “Normal”
If you enjoy math, read this description of a probabilistic model of “normal”…
Human Insight Helps
Finding Anomalies
Once again, if you like math, this description of anomalies is for you…
Take-Home Lesson: Key Steps in Anomaly Detection
A Simple Approach: Threshold Models
Using t-Digest for Threshold Automation
The Philosophy Behind Setting the Threshold
Using t-Digest for Accurate Calculation of Extreme Quantiles
Issues with Simple Thresholds
More Complex, Adaptive Models
Windows and Clusters
Matches with the Windowed Reconstruction: Normal Function
Mismatches with the Windowed Reconstruction: Anomalous Function
A Powerful But Simple Technique
Looking Toward Modeling More Problematic Inputs
Anomalies in Sporadic Events
Counts Don’t Work Well
Arrival Times Are the Key
And Now with the Math…
Event Rate in a Worked Example: Website Traffic Prediction
Extreme Seasonality Effects
No Phishing Allowed!
The Phishing Attack
The No-Phishing-Allowed Anomaly Detector
How the Model Works
Putting It All Together
Anomaly Detection for the Future
Additional Resources