Locally adaptive one-class classifier fusion with dynamic <i>l</i>p-Norm constraints for robust anomaly detection

This paper presents a novel approach to one-class classifier fusion using locally adaptive learning with dynamic lp-norm constraints. Our framework dynamically adjusts fusion weights based on local data characteristics, addressing key challenges in ensemble-based anomaly detection. By incorporating an interior-point optimization technique, our method significantly improves computational efficiency over traditional Frank-Wolfe approaches, achieving up to 19x speed gains in complex scenarios. We evaluate the framework on UCI benchmark datasets and robotics-related temporal sequence datasets, demonstrating superior performance across diverse anomaly types. Statistical validation via Skillings-Mack tests confirms significant advantages over existing methods, consistently achieving top rankings in both pure and non-pure learning scenarios. The framework's ability to adapt to local data patterns while remaining computationally efficient makes it particularly valuable for real-time anomaly detection applications.