The detection method of continuous outliers in complex network data streams based on C-LSTM

Abstract

To enhance the effective detection of abnormal points in complex network data flow, perform multi-dimensional dynamic detection, and establish a more stable and reliable data flow abnormal detection method, a continuous abnormal point detection method for complex network data flow based on C-LSTM is proposed. The features of continuous outliers in complex network data streams are extracted, and a data anomaly detection model is established according to the features. The input features of continuous outliers in complex network data streams are qualitatively and quantitatively transformed into multi-scale anomalies, and the outlier detection based on C-LSTM is realized. The experimental results show that the maximum sensitivity of the proposed method reaches 42%, and the average routing overhead is less than 24 Mb. Regardless of the data in any scenario, the detection accuracy is higher than 0.92, the recall is higher than 0.81, and the F1 value is higher than 0.62. Although there may be some misjudgments or omissions due to noise, the overall detection performance is good.