Yose-Ue: A Treap-Based Ensemble Framework for Resource-Efficient Unsupervised Anomaly Detection

TMLR Paper7985 Authors

Abstract:

Anomaly detection seeks to identify observations that deviate significantly from an underlying data distribution. While deep learning and ensemble-based approaches have achieved strong empirical performance, their computational and memory requirements limit their applicability in resource-constrained edge environments. Furthermore, many approaches to improving efficiency rely on supervised models, which require labeled anomalies that are often scarce in practice. We propose Yose-Ue, a resource-efficient, fully unsupervised anomaly detection framework based on treap-structured ensemble learning. Yose-Ue co-designs a compact data representation with computationally efficient split-selection mechanisms. Specifically, we construct a randomized treap (a hybrid tree–heap data structure) in which nodes are defined by discretized split points, and priorities are assigned via a mass-driven criterion that favors informative partitions. This design yields balanced hierarchical partitions while maintaining low memory overhead. The resulting ensemble estimator improves structural diversity and statistical robustness without incurring the computational cost typical of deep architectures.
We provide a comparative evaluation against established unsupervised ensemble baselines (Isolation Forest, DiForest, and EXTiForest) and resource-efficient state-of-the-art methods, including AutoEncoder, Graph Attention Network AutoEncoder, Histogram-Based Outlier Score, and Local Outlier Factor. Experiments conducted on 14 benchmark datasets—including synthetic datasets, experimental mobile-sensor data, and datasets from the ODDS repository—demonstrate that Yose-Ue achieves competitive or superior detection performance while substantially reducing computational complexity. The proposed method attains over 126× reduction in training time and 7× reduction in inference latency relative to representative baselines. These results indicate that treap-based ensemble learning provides a principled and scalable approach to unsupervised anomaly detection in edge-constrained environments.

Submission Type: Long submission (more than 12 pages of main content)

Assigned Action Editor: ~Chuan-Sheng_Foo1

Submission Number: 7985