Impact of sample size on false alarm and missed detection rates in PCA-based anomaly detection10/04/2017 Objective: Inspiration Principal component analysis based anomaly detection computes the distance of a sample in a dataset to the sample principal subspace of the sample covariance matrix. Points with large distances are labeled as anomalies. Limited training sample size leads to inaccuracy of the sample principal subspace. This work studies how the false alarm and missed detection rates depend on training sample size. Speaker(s)
|