An Anomaly Detection Model Based on Pearson Correlation Coefficient and Gradient Booster Mechanism

Anomaly detection aims to build a decision model that estimates the class of new data based on historical sample features. However, the distance between samples in the feature space is very close sometimes, resulting in samples being invisible to the detection model that is the class overlap problem. To address this issue, an anomaly detection model based on Pearson correlation coefficient and gradient booster mechanism is proposed in this paper. Different from traditional resampling methods, the proposed method groups and sorts features from different dimensions such as feature correlation, feature importance, and feature exclusivity firstly. Then, it selects features with higher correlation and lower importance for deletion to improve the training accuracy of the detector. Furthermore, through the unilateral gradient sampling mechanism, ineffective or inefficient training samples can be further reduced to improve the training efficiency of the detector. Finally, the proposed method was compared with three feature selection methods and six anomaly detection ensemble models on six datasets. The experimental results showed that the proposed method has significant advantages on feature selection, detection performance, detection stability, and computational cost.

@article{Ding2024,
  title     = {An Anomaly Detection Model Based on Pearson Correlation Coefficient and Gradient Booster Mechanism},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {6},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Tuo Ding and He Sui},
  doi       = {10.14569/IJACSA.2024.0150650},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150650}
}