hmu_pca: Multivariate anomaly detector using PCA

In harbinger: A Unified Time Series Event Detection Framework

Multivariate anomaly detector using PCA

Description

Projects multivariate observations onto principal components and flags large reconstruction errors as anomalies. Based on classical PCA.

Usage

hmu_pca()

Details

The series is standardized, PCA is computed, and data are reconstructed from principal components. The reconstruction error is summarized and thresholded.

Value

hmu_pca object.

References

• Jolliffe IT (2002). Principal Component Analysis. Springer.

Examples

library(daltoolbox)

# Load multivariate example data
data(examples_harbinger)

# Use a multidimensional time series
dataset <- examples_harbinger$multidimensional
head(dataset)

# Configure PCA-based anomaly detector
model <- hmu_pca()

# Fit the model (example uses first two columns)
model <- fit(model, dataset[,1:2])

# Run detection
detection <- detect(model, dataset[,1:2])

# Show detected anomalies
print(detection[(detection$event),])

# Evaluate detections
evaluation <- evaluate(model, detection$event, dataset$event)
print(evaluation$confMatrix)

harbinger documentation built on Feb. 11, 2026, 9:06 a.m.