Open Internet by MindsNet

Handling Outlier Detection in Stable Benchmark Data

Outlier detection algorithms like IQR, MAD, and Modified Z-Score can be too aggressive on stable benchmark data, flagging minor normal fluctuations as outliers. This is a known problem called Tight Clustering, where data points are extremely concentrated around the median with minimal dispersion. The goal is to detect 'true anomalies' like OS interruptions, context switches, or garbage collection, not to penalize natural micro variations of a stable system.

Computing & Technology, Computer Science, Machine Learning