This GitHub repository contains the SSIF algorithm, presented in the paper titled Semi-Supervised Isolation Forest for Anomaly Detection
Anomaly detection algorithms attempt to find instances that deviate from the expected behavior. Because this is often tackled as an unsupervised task, anomaly detection models rely on exploiting intuitions about what constitutes anomalous behavior. These typically take the form of data-driven heuristics that measure the anomalousness of each instance. However, the unsupervised detectors are limited by the validity of their intuition. Because these are not universally true, one can improve the detectors’ performance by using a semi-supervised approach that exploits a few labeled instances. This paper proposes a novel semi-supervised tree ensemble based anomaly detection framework. We compare our proposed approach to several baselines and show that it performs comparably well to the best state-of-the-art neural networks on 6 real-world and 14 benchmark datasets.
The folder contains:
- SSIF.py, the file containing the model proposed in this paper;
- Notebook.ipynb, a notebook showing how to train SSIF and compute the anomaly scores on an artificial dataset;
- Supplement.pdf, a pdf with the paper's supplementary material.
Given a training dataset X_train with training labels Y_train(where y=0 for the unlabeled data, y=1 for the anomalies and y=-1 for the normals), and a test dataset X_test, the algorithm is applied as follows:
from ssif import SSIF
from sklearn.metrics import roc_auc_score
# Train the model
detector = SSIF()
detector.fit(Xtrain, ytrain)
# Compute the anomaly scores for the training set
train_scores_ssif = detector.compute_anomaly_scores(X_train)
# Compute the anomaly scores for the test set
test_scores_ssif = detector.compute_anomaly_scores(X_test)
# Compute the AUROC performance of the model
auc_performance = roc_auc_score(Y_test, test_scores_ssif)This folder requires the following python packages to be used: