Outlier Ensemble Based on Isolation Forest: The CBOEA Approach
Open Access
- 1 February 2025
- journal article
- Published by Walter de Gruyter GmbH in Foundations of Computing and Decision Sciences
- Vol. 50 (1) , 27-55
Abstract
Outliers are instances that deviate from the norm. In certain fields, their detection is crucial since they are often indicators of interesting events such as system faults and deliberate human actions. Anomaly detection is an essential data mining task that is employed in many real-life applications. The continuous development of anomaly detection algorithms is primarily motivated by the explosive growth in both size and number of attributes of the data sets. Such growth requires algorithms that can deal with large data sets with e↵ectiveness and efficiency. Isolation Forest (IF) was introduced with that idea in mind. IF uses an isolation mechanism to detect outliers without relying on any distance or density measures. This approach handles large data sets quite well, thanks to its low time complexity. However, IF struggles to detect local outliers. In this work, a new algorithm called Cluster-Based Outlier Ensemble Approach (CBOEA) is proposed. This approach combines IF and Local Outlier Factor (LOF) outputs through a clustering algorithm called OPTICS to identify the clustering structure. This clustering technique allows the compensation of IF weaknesses while maintaining its strengths. The proposed algorithm is then compared to LOF and IF using two evaluation metrics. The performance with benchmark data sets shows that the proposed method is competitive with its components.Keywords
This publication has 28 references indexed in Scilit:
- A novel anomaly score based on kernel density fluctuation factor for improving the local and clustered anomalies detection of isolation forestsInformation Sciences, 2023
- TiWS-iForest: Isolation forest in weakly supervised and tiny ML scenariosInformation Sciences, 2022
- Anomaly explanation: A reviewData & Knowledge Engineering, 2022
- A probabilistic generalization of isolation forestInformation Sciences, 2022
- Enhanced anomaly scores for isolation forestsPattern Recognition, 2021
- Anomaly Detection2017
- Outlier Ensembles2017
- Outlier detection in audit logs for application systemsInformation Systems, 2014
- Improving iForest with Relative Mass2014
- ST-DBSCAN: An algorithm for clustering spatial–temporal dataData & Knowledge Engineering, 2007