Title: An outlier detection method for open-source software running data based on Bi-LSTM network

Authors: Jiehai Deng; Weihong Li

Addresses: JiangXi College of Traditional Chinese Medicine, Fuzhou, Jiangxi Province, China ' Science and Technology Project Service Centre of Fuzhou, Fuzhou, Jiangxi Province, China

Abstract: Aiming at the problem of significant errors in existing data outlier detection methods, a Bi-LSTM network-based open-source software running data outlier detection method is proposed. This method combines parametric and non-parametric methods, sets appropriate threshold values for outlier features in open-source software running data, and extracts features using Manhattan distance and outlier factors in the K-nearest neighbour algorithm. The XGBoost algorithm is introduced to define a decision tree and train leaf nodes to obtain classification features by reducing the loss function of classification residuals. By standardising and normalising the obtained features, combined with the Bi-LSTM network algorithm and attention mechanism, data outlier detection is carried out. The experimental results show that this method has achieved better results in anomaly detection.

Keywords: Bi-LSTM network; open-source software; operating data; abnormal value detection; Manhattan distance; forward detection; reverse detection.

DOI: 10.1504/IJCAT.2024.143297

International Journal of Computer Applications in Technology, 2024 Vol.74 No.4, pp.247 - 257

Received: 19 Jan 2024
Accepted: 30 Apr 2024
Published online: 12 Dec 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article