Integrating SHAP and Attention Mechanisms for Transparent Deep Autoencoder Anomaly Detection

Telu, Venkateshwar Rao (2025) Integrating SHAP and Attention Mechanisms for Transparent Deep Autoencoder Anomaly Detection. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (999kB) | Preview
Preview	PDF (Configuration Manual) Download (243kB) | Preview

Abstract

This study aims to develop and evaluate a transparent deep autoencoder anomaly detection framework by integrating SHAP-based feature attribution and attention mechanisms, there by enhancing interpretability while maintaining high detection performance in network traffic contexts. This study follows the CRISP-DM framework, establishing a correspondence between the phases and the anomaly detection pipeline. The Network Anomaly Dataset has been collected in the form of a CSV on the Kaggle site, and the data has been loaded into Pandas using read into CSV. The adaptive combination of the information latencies across the channels and along time axes was realized by the interleaving blocks of CBAM in consecutive convolutional layers, and the more basic latent representations were realized by Multi Head Attention. Additionally, perfectly tuned Isolation Forest demonstrated an extremely elevated level of outlier scoring of sparse anomalies in fewer splits, which functioned as a good benchmark. Post-hoc SHAP interpretations identified that session state, destination host metrics, and connection flag patterns were the key factors influencing anomalous reconstruction, and most auxiliary features were minor contributors.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Singh, Jaswinder UNSPECIFIED
Uncontrolled Keywords:	SHAP-based feature attribution; Network Anomaly; CBAM; Autoencoder; Isolation Forest
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QA Mathematics > Computer software > Computer Security T Technology > T Technology (General) > Information Technology > Computer software > Computer Security Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	27 Aug 2026 09:03
Last Modified:	27 Aug 2026 09:03
URI:	https://norma.ncirl.ie/id/eprint/9677

Actions (login required)

View Item