Anomaly Identification and Prediction Method for Network Timing Data Based on Improved Isolated Forest Approach

Aiming at the problem of insufficient real-time and poor adaptability to dynamic changes in network timing data anomaly detection, this paper proposes an anomaly identification and prediction method based on temporal convolutional network and isolated forest. The method firstly uses time convolution network to extract features from network timing data, effectively capturing the timing dependence and long-term trend in the data. Subsequently, the isolated forest algorithm is used for fast anomaly identification and prediction of the extracted features, and the accuracy and efficiency of anomaly detection is improved by optimising the hyperparameters of the isolated forest. Experiments demonstrate that the method proposed in this paper can significantly improve the accuracy and real-time performance of anomaly detection compared with the traditional isolated forest method, and provide lightweight and adaptive anomaly warning support for dynamic network environments.