Machine Learning Tools – FAQ

Overview

Seeq provides built-in Machine Learning Tools that enable engineers and analysts to build soft sensors, detect anomalies, and integrate custom models.

• Prediction Models – estimate signal values from multivariate inputs:

  • Partial Least Squares (PLS)

  • External ML (ONNX-based)

• Anomaly Detection Models – identify unusual operating regimes, correlations, or outliers:

  • Self-Organizing Maps (SOM)

  • Isolation Forest (iForest)

  • Principal Component Analysis (PCA)

  • Clustering

  • External ML (ONNX-based)

General FAQs

Q: What types of ML models does Seeq support?
A: Prediction models (PLS, External ML) and anomaly detection models (SOM, Isolation Forest (iForest), PCA, Clustering, External ML).

Q: What input data can I use?
A: Numeric time-series signals. Categorical or string data is not supported.

Q: How is the training window defined?
A: By default, the tool uses the time range set in the display pane. Customers can optionally limit training to a Condition within that time frame.

Q: Do all tools require a training window?
A: Yes. A representative training window is needed to establish baseline behavior or relationships.

Q: What outputs do these models generate?
A:

• Prediction tools (PLS, External ML) → Predicted signal(s) and error statistics.

• Anomaly detection tools (SOM, iForest, PCA, Clustering, External ML) → Anomaly condition(s) and/or score signals.

Q: Can models update automatically?
A: It depends on how the training window is defined:

• Fixed training window (display range): The model does not update automatically; you must retrain if you want new data included.

• Condition-based training window: If the model is limited to a Condition, and that condition updates with new data, the model will automatically retrain using the updated condition.

Q: How do I choose between anomaly detection tools?
A:

• SOM → Unusual operating regimes.

• iForest → Rare, isolated outliers.

• PCA → Deviations in correlation structure.

• Clustering → Groups of similar operating states.

• External ML → Custom anomaly detection model.

Q: How do I choose between prediction tools?
A:

• PLS → Built-in regression for predicting an output from input signals.

• External ML → When you already have a predictive model (e.g., regression, classification) trained outside Seeq.

Q: What are common use cases?
A:

• SOM: Detecting regime shifts.

• iForest: Detecting outliers.

• PCA: Monitoring process health.

• Clustering: Identifying operating groups.

• PLS: Building soft sensors.

• External ML: Deploying custom ONNX models for prediction or anomaly detection.

Q: Do I need a premium license?
A: External ML requires a separate license. All other ML tools are included in the standard package. If you are interested in External ML please contact your Customer Service Mananger or Sales Exec for more information.

Q: Where can I find detailed instructions?
A: Each tool has a Knowledge Base article: SOM, Isolation Forest, PLS, PCA, Clustering, External ML.

Which ML Tool Should I Use? – Decision Guide

What is your goal?

• I want to predict a variable (soft sensor, quality, future value):

  • Use PLS if you want a built-in regression method that relates multiple input (X) signals to a target (Y).

  • Use External ML (ONNX) if you already have a predictive model in ONNX format. These models are usually developed in external machine learning environments or workflows, but can also be created and exported from Seeq Data Lab.

• I want to detect anomalies (outliers, unusual regimes, deviations):

  • Use SOM to detect unusual operating regimes by comparing new states to learned clusters.

  • Use iForest to detect rare, isolated outliers.

  • Use PCA to catch deviations in multivariate correlation structure.

  • Use Clustering to group similar operating states and flag atypical periods.

  • Use External ML (ONNX) if you already have an anomaly detection model in ONNX format. As with predictive models, these are generally built in external ML environments or workflows, but can also be developed in Seeq Data Lab and exported to ONNX.

ML Tool Comparison Table