Randomized PCA Forest for Unsupervised Outlier Detection
Researchers have proposed a novel unsupervised outlier detection method based on Randomized Principal Component Analysis (PCA). Motivated by the strong performance of Randomized PCA (RPCA) Forest in approximate K-Nearest Neighbor (KNN) search, the team developed a new approach that derives an outlier score from the intrinsic properties of the RPCA Forest. This method aims to enhance the accuracy and efficiency of identifying anomalies in datasets without labeled data. Experimental results demonstrate that the proposed approach outperforms classical and state-of-the-art methods on several datasets while remaining competitive on others. Extensive analysis highlights the method's robustness and computational efficiency, positioning it as a highly effective choice for unsupervised outlier detection tasks. The paper, authored by Muhammad Rajabinasab, Farhad Pakdaman, Moncef Gabbouj, Peter Schneider-Kamp, and Arthur Zimek, was submitted to arXiv under the Computer Science > Machine Learning category. It represents a significant advancement in machine learning techniques for anomaly detection, offering a scalable solution for complex data analysis challenges.
Wire timeline
Randomized PCA Forest for Unsupervised Outlier Detection
Researchers have proposed a novel unsupervised outlier detection method based on Randomized Principal Component Analysis (PCA). Motivated by the strong performance of Randomized PCA (RPCA) Forest in approximate K-Nearest Neighbor (KNN) search, the team developed a new approach that derives an outlier score from the intrinsic properties of the RPCA Forest. This method aims to enhance the accuracy and efficiency of identifying anomalies in datasets without labeled data. Experimental results demonstrate that the proposed approach outperforms classical and state-of-the-art methods on several datasets while remaining competitive on others. Extensive analysis highlights the method's robustness and computational efficiency, positioning it as a highly effective choice for unsupervised outlier detection tasks. The paper, authored by Muhammad Rajabinasab, Farhad Pakdaman, Moncef Gabbouj, Peter Schneider-Kamp, and Arthur Zimek, was submitted to arXiv under the Computer Science > Machine Learning category. It represents a significant advancement in machine learning techniques for anomaly detection, offering a scalable solution for complex data analysis challenges.
cs.AI updates on arXiv.org