基于RAdaboost-SVDD的异常值检测OA

Support Vector Data Description(SVDD)is widely used in anomaly detection for one-class classification(OCC).However,its performance may be limited when dealing with complex data distributions or high-noise scenarios.To address this issue,this paper proposes an improved method RAdaboost-SVDD that combines Adaboost and SVDD.By using SVDD as a weak classifier,multiple SVDD models are constructed under the Adaboost framework to form an integrated strong classifier.The ini-tial hypersphere is built using SVDD,and during each iteration,sample weights are dynamically adjusted to enhance the model's focus on misclassified samples,thereby progressively improving overall detection performance.Finally,the strong classifier is in-tegrated using a voting mechanism.Although SVDD itself is a strong classifier with high sensitivity to anomalies,its combination with traditional Adaboost can lead to significant amplification of misclassification losses for anomalies due to Adaboost's use of an exponential loss function.This over-sensitivity to anomalies increases the overall training error.To address this issue,this pa-per modifies the Adaboost loss function to reduce overemphasis on anomalies,thereby improving the model's robustness.Experi-mental results demonstrate that the proposed method significantly improves anomaly detection accuracy and robustness when han-dling noisy and complex data distributions and fully verifies that it has advantages and practicality in the single-class classifica-tion task.