Principal Component Analysis Outputs

The following outputs can be generated by the Principal Component Analysis tool. Depending on your configuration and analysis requirements, you can select one or more outputs from the Outputs section.

Ouptut	Type	Description
Anomalies	Condition	A condition that identifies observations classified as anomalous by the PCA model based on the configured thresholds.
Log-Likelihood Signal	Signal	A monitoring signal that uses a log-likelihood method to indicate deviations from the expected behavior.
T² Signal	Signal	A type of monitoring signal which measures the distance from the mean in the principal component space.
SPE Signal	Signal	A type of monitoring signal which measures the distance from the principal component space
Anomaly Threshold	Scalar	Displays the anomaly threshold specified during model configuration. For example, a value of 99% indicates that observations outside the expected range at the configured confidence level may be identified as anomalies.
Cumulative R-Squared	Scalar	Shows the cumulative amount of variation in the training data explained by the selected principal components.
Eigenvalues	Scalar	Indicates the amount of variation captured by each principal component. Larger values indicate more influential components.
Explained Variance Ratio	Scalar	Shows the percentage of total variation explained by each principal component.
Hotelling T² Limit	Scalar	The upper limit used to evaluate the T² signal. Observations exceeding this limit may indicate unusual behavior.
Log-Likelihood Limit	Scalar	The lower limit used to evaluate the log-likelihood signal. Observations beyond this limit may indicate unusual behavior.
Mean Squared Error	Scalar	Measures the average reconstruction error of the PCA model on the training data. Lower values generally indicate a better representation of the training data.
Model input properties	Scalar	Provides information about the model inputs used during training. This output can be used for model validation and troubleshooting.
Monitoring Signal Contrast	Scalar	Indicates the relative contribution of the monitoring signals used for anomaly detection. This output can help assess how strongly observations differ from normal operating behavior.
Number of Principal Components	Scalar	Shows the number of principal components specified during model training.
PCA Loadings	Scalar	Shows how strongly each input signal contributes to each principal component. This output can help identify which variables influence the model most.
SPE Limit	Scalar	The upper limit used to evaluate the SPE signal. Observations exceeding this limit may indicate abnormal behavior.
Total Explained Variance	Scalar	Shows the total percentage of variation in the training data explained by the selected principal components.
Training window	Condition	Identifies the time range used to train the Principal Component Analysis model. This output can be used to verify the training period and compare it with scored data.