User and Entity Behavior Analysis under Urban Big Data

TLDR

Rapid expansion of urban network infrastructure is turning cities into smart cities, but the proliferation of smart devices collecting private data creates serious security risks. This study aims to detect insiders’ anomalous behaviors to prevent urban big data leakage. The authors model users’ daily activities from four perspectives, compute deviations using LSTM and convLSTM, and classify anomalies with an MLP, evaluating the multimodel-based system on the CERT dataset. Experiments demonstrate that the multimodel-based system effectively learns normal activity patterns and accurately detects anomalous behaviors.

Abstract

Recently, the urban network infrastructure has undergone a rapid expansion that is increasingly generating a large quantity of data and transforming our cities into smart cities. However, serious security problems arise with this development with more and more smart devices collecting private information under smart city scenario. In this article, we investigate the task of detecting insiders’ anomalous behaviors to prevent urban big data leakage. Specifically, we characterize a user's daily activities from four perspectives and use several deep learning algorithms (long short-term memory (LSTM) and convolutional LSTM (convLSTM)) to calculate deviations between realistic actions and normalcy of daily behaviors and use multilayer perceptron (MLP) to identify abnormal behaviors according to those deviations. To evaluate the proposed multimodel-based system (MBS), we conducted experiments on the CERT (United States Computer Emergency Readiness Team) dataset. The experimental results show that our proposed MBS has a remarkable ability to learn the normal pattern of users’ daily activities and detect anomalous behaviors.