Multivariate Process Control by Principal Component Analysis Using T² and Q errors

Using and interpreting Hotelling’s T² and Squared Prediction Error Q in anomaly detection systems

A substantial part of my job as a data scientist consists in building anomaly detection systems for process control in manufacturing, and the Principal Component Analysis (from now PCA) has an essential role in my toolbox.

The scientific literature suggests two measures to trace anomalies through the PCA: Hotelling’s T² and Squared Prediction Error, aka Q error. Although their calculation …

analysis anomaly anomaly detection building control data data scientist detection error errors job literature manufacturing multivariate multivariate-analysis part pca-analysis prediction principal-component process role software systems