KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Anomaly
Observation or pattern that deviates significantly from expected or normal behavior in a dataset, potentially indicating an error, fraud, or a rare event.
Outlier
Data point that differs considerably from other observations, often identified by statistical methods or unsupervised machine learning algorithms.
Isolation Forest
Anomaly detection algorithm based on decision trees that isolates anomalies by randomly building forests of trees and measuring the average isolation depth of points.
Local Outlier Factor (LOF)
Local density-based anomaly detection method that compares the density of a point with that of its neighbors to identify points in low-density regions.
One-Class SVM
Variant of support vector machines trained on normal data to create a decision boundary, identifying as anomalies the points located outside this boundary.
Autoencoder
Unsupervised neural network learned to reconstruct its input data, where anomalies are identified by high reconstruction error indicating a deviation from the normal pattern.
Elliptic Envelope
Robust statistical method that fits an ellipse around normal data using robust covariance estimation, considering outside points as anomalies.
Z-score
Standardized statistical measure that quantifies the deviation of an observation from the mean in units of standard deviation, with extreme values potentially indicating anomalies.
IQR (Interquartile Range)
Detection method based on the range between the first and third quartiles, where observations deviating by 1.5 times the IQR beyond the quartiles are considered anomalies.
Mahalanobis Distance
Multivariate distance measure that takes into account the correlation between variables, identifying anomalies as points with a high distance from the center of the distribution.
Reconstruction Error
Difference between the original data and their reconstruction by a model like an autoencoder, where high errors indicate abnormal observations.
Novelty Detection
Variant of anomaly detection where the model is trained on normal data to identify new unknown observations that deviate from learned patterns.
Point Anomaly
Type of anomaly where an individual observation is considered abnormal compared to the rest of the data, without depending on context or other observations.
Contextual Anomaly
Anomaly identified as abnormal only in a specific context, such as a high temperature in winter that might be normal in summer.
Collective Anomaly
Group of observations that, individually normal, become abnormal when they appear together as a sequential or spatial collection.
Robust Covariance
Covariance estimation method resistant to extreme values, used to detect anomalies by identifying points that significantly deviate from the robust distribution.
Clustering-based Anomaly Detection
Approach that uses clustering algorithms to identify anomalies as points not belonging to any cluster or belonging to very small clusters.
Histogram-based Outlier Detection
Method that builds multidimensional histograms of data and identifies anomalies as observations falling into bins with very low frequencies.