异常检测

检测离群值和新奇值

Statistics and Machine Learning Toolbox™ 为未标注的多元样本数据提供了几种异常检测方法。您还可以使用一类支持向量机 (SVM) 和稳健的随机切割森林算法（请参阅Incremental Anomaly Detection Overview）来检测流数据中的异常。异常检测通过训练模型或学习参数来检测离群值（训练数据中的异常）。对于新奇值检测（使用未污染的训练数据检测新数据中的异常），您可以使用未污染的训练数据（没有离群值的数据）训练模型或学习参数，并通过使用训练的模型或学习的参数来检测新数据中的异常。有关详细信息，请参阅Unsupervised Anomaly Detection。

如果将训练数据标注为正常点和异常点，您可以训练二类分类模型，并使用 resubPredict 和 predict 对象函数分别检测训练数据和新数据中的异常。有关支持的分类方法列表，请参阅分类。

该工具箱还提供了特定于模型的异常检测方法，您可以在训练分类、回归或聚类模型后应用这些方法。有关详细信息，请参阅 Model-Specific Anomaly Detection。

函数

全部展开

孤立森林

`iforest`	Fit isolation forest for anomaly detection (自 R2021b 起)
`isanomaly`	Find anomalies in data using isolation forest (自 R2021b 起)

稳健随机切割森林

`rrcforest`	Fit robust random cut forest model for anomaly detection (自 R2023a 起)
`isanomaly`	Find anomalies in data using robust random cut forest (自 R2023a 起)

局部离群值因子

`lof`	Create local outlier factor model for anomaly detection (自 R2022b 起)
`isanomaly`	Find anomalies in data using local outlier factor (自 R2022b 起)

一类支持向量机 (SVM)

`ocsvm`	Fit one-class support vector machine (SVM) model for anomaly detection (自 R2022b 起)
`isanomaly`	Find anomalies in data using one-class support vector machine (SVM) (自 R2022b 起)

马氏距离

`robustcov`	Robust multivariate covariance and mean estimate
`mahal`	Mahalanobis distance to reference samples
`pdist2`	两组观测值之间的两两距离

增量稳健随机切割森林

`incrementalLearner`	Convert robust random cut forest model to incremental learner (自 R2023b 起)
`fit`	Train robust random cut forest model for incremental anomaly detection (自 R2023b 起)
`isanomaly`	Find anomalies in data using robust random cut forest (RRCF) for incremental learning (自 R2023b 起)
`reset`	Reset incremental robust random cut forest model (自 R2023b 起)

增量一类支持向量机 (SVM)

`incrementalLearner`	Convert one-class SVM model to incremental learner (自 R2023b 起)
`fit`	Train one-class SVM model for incremental anomaly detection (自 R2023b 起)
`isanomaly`	Find anomalies in data using one-class support vector machine (SVM) for incremental learning (自 R2023b 起)
`reset`	Reset incremental one-class SVM model (自 R2023b 起)

对象

`IsolationForest`	Isolation forest for anomaly detection (自 R2021b 起)
`RobustRandomCutForest`	Robust random cut forest model for anomaly detection (自 R2023a 起)
`LocalOutlierFactor`	Local outlier factor model for anomaly detection (自 R2022b 起)
`OneClassSVM`	One-class support vector machine (SVM) for anomaly detection (自 R2022b 起)
`incrementalRobustRandomCutForest`	Robust random cut forest model for incremental anomaly detection (自 R2023b 起)
`incrementalOneClassSVM`	One-class support vector machine (SVM) model for incremental anomaly detection (自 R2023b 起)

主题

Unsupervised Anomaly Detection
Detect anomalies using isolation forest, robust random cut forest, local outlier factor, one-class SVM, and Mahalanobis distance.
Incremental Anomaly Detection Overview
Discover fundamental concepts about incremental anomaly detection in streaming data.
Configure Model for Incremental Anomaly Detection
Prepare a model for incremental anomaly detection and fitting on a data stream.
Anomaly Detection with Isolation Forest
Detect anomalies by isolating anomalies from normal points using an isolation forest (ensemble of isolation trees).
Model-Specific Anomaly Detection
After training a classification, regression, or clustering model, detect anomalies using a model-specific anomaly detection feature.
Build Condition Model for Industrial Machinery and Manufacturing Processes
Train a binary classification model using Classification Learner App to detect anomalies in sensor data collected from an industrial manufacturing machine.
Code Generation for Anomaly Detection
Generate single-precision code that detects anomalies in data using a trained isolation forest model or one-class SVM.