Anomaly Detection Algorithm Based on Broad Learning System and Support Vector Domain Description

Deep neural network-based autoencoders can effectively extract high-level abstract features with outstanding generalization performance but suffer from sparsity of extracted features, insufficient robustness, greedy training each layer, and a lack global optimization. In this study, the broad learning system (BLS) is improved to obtain new model for data reconstruction. Support Vector Domain Description (SVDD) one best-known one-class-classification methods used solve problems where proportion sample categories extremely unbalanced. The SVDD sensitive penalty parameters C, which represents trade-off between sphere volume number target outside sphere. process only considers normal samples, leads low recall rate weak performance. To address these issues, we propose BLS-based weighted algorithm (BLSW_SVDD), introduces reconstruction error weights small anomalous samples when model, thus improving robustness model. evaluate BLSW_SVDD comparison experiments were conducted on UCI dataset, experimental results showed that in terms accuracy F1 values, has better advantages than traditional algorithms.