ConXsense - Automated Context Classification for Context-Aware Access Control

TL;DR

ConXsense is a novel framework that uses machine learning and context sensing to automatically classify smartphone contexts for improved, user-preferred access control and security, addressing limitations of prior static or user-defined policies.

Contribution

It introduces a probabilistic, context-aware access control framework that automatically classifies security-related contexts on mobile devices, integrating with Android for real-time enforcement.

Findings

Effective context classification using real-world data

Improved security against device misuse and sensory malware

User perceptions align with framework's security assessments

Abstract

We present ConXsense, the first framework for context-aware access control on mobile devices based on context classification. Previous context-aware access control systems often require users to laboriously specify detailed policies or they rely on pre-defined policies not adequately reflecting the true preferences of users. We present the design and implementation of a context-aware framework that uses a probabilistic approach to overcome these deficiencies. The framework utilizes context sensing and machine learning to automatically classify contexts according to their security and privacy-related properties. We apply the framework to two important smartphone-related use cases: protection against device misuse using a dynamic device lock and protection against sensory malware. We ground our analysis on a sociological survey examining the perceptions and concerns of users related to contextual smartphone security and analyze the effectiveness of our approach with real-world context data. We also demonstrate the integration of our framework with the FlaskDroid architecture for fine-grained access control enforcement on the Android platform.