AI 용어집
인공지능 완전 사전
Contamination
Hyperparameter of the Isolation Forest defining the expected proportion of anomalies in the dataset, used to calibrate the decision threshold on anomaly scores.
iTree (Isolation Tree)
Binary decision tree randomly built in an Isolation Forest, where each partition is performed on a randomly selected feature and split value until each point is isolated.
Isolation
Principle stating that anomalies are 'few and different' data points, and are therefore easier to isolate than normal points, requiring fewer partitions in a tree.
Extension Level
Optional hyperparameter in some Isolation Forest implementations that controls the number of dimensions to split simultaneously at each node, allowing handling of very high-dimensional data.
Scoring Function
Mathematical function that transforms the average path length into a normalized anomaly score, typically based on the distribution of path length in a random tree.
Forest Size
Number of iTrees built in the Isolation Forest, a hyperparameter that influences the stability and accuracy of the final anomaly scores.
Node Depth
Depth of a specific node in an iTree, used to calculate the path length and thus the isolation score of a data point.
External Anomalies
Type of anomalies that lie outside regions of high data density, effectively detected by the Isolation Forest due to their rapid isolation.
Computational Complexity
Measure of the efficiency of the Isolation Forest, which is linear with respect to the number of observations and sublinear with respect to the number of features, making it suitable for large datasets.