A Requirement-Based Framework for Engineering Adaptive Authentication

1. Problem Statement

In dynamic environments such as the Internet of Vehicles (IoV), healthcare, and smart cities, security risks and system requirements (e.g., usability, performance) fluctuate based on contextual factors (e.g., location, time of day, device type, network topology).

• Limitations of Static Authentication: Traditional static authentication methods (relying on fixed credentials) fail to adapt to these changing risks. They often either over-secure a low-risk scenario (hurting usability/performance) or under-secure a high-risk scenario (leaving the system vulnerable).
• Research Gaps: Existing adaptive authentication systems lack:
  1. Formal representations of contextual factors, requirements, and authentication methods.
  2. Mechanisms to understand how context influences the prioritization of requirements and the feasibility of specific methods.
  3. Automated techniques to select the optimal authentication method that balances risk mitigation with requirement satisfaction.

2. Methodology

The authors propose a Requirement-Based Framework grounded in the MAPE-K (Monitor, Analyze, Plan, Execute, Knowledge) self-adaptive loop. The core innovation lies in the Knowledge representation and the Analyze/Plan decision engine.

A. Knowledge Representation

The framework models the system using three interconnected components:

1. Contextual Goal Model:

   • Represents system requirements (Security: Confidentiality, Integrity, Authenticity; Usability: Effectiveness, Efficiency; Performance).
   • Maps Contextual Factors (e.g., "Ambulance at Junction," "Low Lighting") to the priority of these goals.
   • Uses a quantitative scale $[0, 1]$ to represent the impact of context on goal priority (e.g., $>0.8$ is "very strong positive").
   • Aggregates multiple context impacts using the maximum operator to ensure a conservative, risk-aware prioritization.

2. Extended Feature Model:

   • Represents the spectrum of available Authentication Methods (e.g., Password, Biometrics, Smartcards, Two-Factor).
   • Defines features such as credential type, automation level, and cryptographic strength.
   • Explicitly models feasibility constraints: Certain contexts render methods infeasible (e.g., "Low Lighting" $\rightarrow$ Face/Iris recognition = 0).
   • Quantifies the contribution of each method to satisfying specific requirements (e.g., Certificate-based auth has high impact on Authenticity but negative impact on Performance).

3. Security Risk Model:

   • Calculates risk based on Likelihood (influenced by context and enabled methods) and Harm (influenced by data sensitivity).
   • Uses a conservative aggregation strategy: Total Risk = $\max(\text{Partial Risks})$.
   • Authentication methods reduce the likelihood of attacks (e.g., Two-Factor reduces impersonation risk significantly).

B. Decision Engine: Fuzzy Causal Network (FCN) & Z3

To dynamically select the best authentication method, the framework employs a Fuzzy Causal Network (FCN) encoded using the Z3 SMT (Satisfiability Modulo Theories) solver.

• FCN Structure:
  • Chance Nodes: Contextual factors and goals.
  • Decision Nodes: Binary selection of authentication configurations (0 or 1).
  • Utility Node: A metric to maximize, defined as the average of goal satisfactions plus the complement of total risk:
    $$Utility = \frac{S(Security) + S(Usability) + S(Performance) + (1 - TotalRisk)}{4}$$
• Execution:
  • At runtime, the system monitors context and updates the FCN constraints.
  • Z3 performs a binary search over the utility interval $[0, 1]$ to find the authentication configuration that maximizes utility while satisfying all feasibility constraints.

3. Key Contributions

1. Unified Modeling Framework: A novel integration of Contextual Goal Models and Extended Feature Models to formally represent the interdependencies between context, requirements, and authentication mechanisms.
2. Automated Runtime Adaptation: A technique using Z3-encoded FCNs to automatically compute the optimal authentication method in response to changing contexts, balancing conflicting goals (e.g., Security vs. Usability).
3. Feasibility Handling: Explicit modeling of how environmental factors (e.g., lighting, network type) disable specific authentication methods, preventing the selection of ineffective solutions.
4. Quantitative Risk-Utility Trade-off: A formal utility function that mathematically balances the reduction of security risks against the satisfaction of non-functional requirements.

4. Results and Evaluation

The framework was evaluated against a static baseline (enforcing a single authentication method across all scenarios) in two distinct domains: Internet of Vehicles (IoV) and Healthcare.

Case Study A: IoV Scenarios

• Scenarios: Secure Data Exchange (V2R), Ambulance Overtaking (V2V), Accessing Patient Data at a Junction (V2I).
• Findings:
  • The static baseline failed to consistently satisfy requirements. For example, using certificates for V2V overtaking (high performance need) caused unacceptable delays.
  • The proposed framework dynamically selected:
    • Certificate-based auth for V2R (High Security priority).
    • Car Plate/License for V2V (High Performance priority).
    • Two-Factor (Plate + Fingerprint) for V2I (Balanced Security/Usability).
  • Overhead: Runtime analysis took < 1.5 seconds (avg ~960ms) with < 19 MB memory usage, proving feasibility for runtime deployment.

Case Study B: Healthcare Scenarios

• Scenarios: Prescribing meds outside hospital (S4), Emergency Dept access (S5), Late-night home access (S6).
• Findings:
  • The framework successfully adapted to constraints (e.g., disabling biometrics in "Low Lighting" or "No Scanner" scenarios).
  • It selected Two-Factor (SmartCard + Fingerprint) for high-risk external access (S4) and SmartCard only for high-performance emergency access (S5).
  • Overhead: Execution times ranged from 358ms to 3007ms depending on the complexity of the search space (fewer enabled methods = faster resolution). Memory usage remained stable (~10-15 MB).

Comparison with Baseline

In both domains, the adaptive framework consistently achieved higher utility and lower total security risk compared to the static baseline. The baseline often failed to mitigate risks in high-threat scenarios or degraded performance in low-threat scenarios.

5. Significance

• Bridging Design and Runtime: The work bridges the gap between design-time modeling (Goal/Feature models) and runtime decision-making, enabling truly self-adaptive security systems.
• Context-Aware Security: It moves beyond simple "risk scoring" to a holistic approach that considers the feasibility of methods and the trade-offs between security, usability, and performance.
• Scalability: The use of Z3 demonstrates that complex constraint solving for adaptive security is computationally feasible for real-time applications, even in resource-constrained environments like IoV.
• Generalizability: By successfully applying the framework to two disparate domains (transportation and healthcare), the authors demonstrate its potential applicability to other dynamic environments like Smart Cities and Industrial IoT.

Conclusion

The paper presents a robust, mathematically grounded framework for engineering adaptive authentication. By leveraging formal models and SMT solvers, it enables systems to dynamically select authentication methods that maximize utility in the face of changing security risks and operational constraints, offering a significant advancement over static or heuristic-based adaptive approaches.