Statistics and Machine Learning Toolbox™ provides several anomaly detection features for unlabeled multivariate sample data. The anomaly detection features detect outliers (anomalies in training data) either by training a model or by learning parameters. For novelty detection (detecting anomalies in new data with uncontaminated training data), you train a model or learn parameters with uncontaminated training data (data with no outliers) and detect anomalies in new data by using the trained model or learned parameters. For more details, see Unsupervised Anomaly Detection.
If you have training data labeled as normal points and anomalies, you can
train a binary classification model and use the
functions to detect anomalies in the training data and new data, respectively.
For the list of supported classification features, see Classification.
The toolbox also provides model-specific anomaly detection features that you can apply after training a classification, regression, or clustering model. For details, see Model-Specific Anomaly Detection.
Detect anomalies using isolation forest, one-class support vector machine (OCSVM), and Mahalanobis distance.
Detect anomalies by isolating anomalies from normal points using an isolation forest (ensemble of isolation trees).
After training a classification, regression, or clustering model, detect anomalies using a model-specific anomaly detection feature.