False Data Detection for Fog and Internet of Things Networks

Romano Fantacci, Francesca Nizzi, Tommaso Pecorella, Laura Pierucci, Manuel RoveriSensors 2019 · OpenReview

TL;DR

This paper introduces DataIDS, a novel intrusion detection system for Fog and IoT networks that leverages real-time analysis of sensed data to detect and isolate malicious data injection attacks. By utilizing Fog Computing for enhanced processing and a dependency graph to model sensor relationships, DataIDS offers a more robust and timely security solution compared to traditional network-level intrusion detection systems.

What Problem Does This Paper Address?

The proliferation of Internet of Things (IoT) devices, while enabling unprecedented connectivity and data generation, introduces significant security vulnerabilities. These resource-constrained devices are often susceptible to attacks, and their sheer number and heterogeneity make traditional security approaches challenging. A key concern is the injection of false data into the system, which can lead to incorrect decision-making and system malfunctions. Existing intrusion detection systems (IDS) often focus on network traffic or system logs, which may not be sufficient to detect sophisticated attacks targeting the actual sensor data in a Fog/IoT environment. The paper addresses this gap by proposing a data-centric approach to detect and isolate these malicious data injections.

Methods

The core of the proposed solution is the Data Intrusion Detection System (DataIDS), implemented within a Fog Computing (FC) framework. Here's a breakdown of the key methods:

  • Fog Computing Integration: DataIDS is deployed on Fog nodes, which are positioned closer to the IoT devices than the cloud. This allows for faster data processing, reduced network overhead, and the ability to leverage local information (like network topology) for enhanced security.
  • Dependency Graph: A novel approach involves constructing a "dependency graph" where nodes represent sensors and edges represent the cross-correlation between their data streams. This graph captures the functional relationships between sensors, even if they are heterogeneous (e.g., measuring temperature and humidity). The graph is built using the normalized absolute value of the peak cross-correlation between data sequences, with a user-defined threshold ($\gamma$) to define an edge.
  • Change Detection: Once the dependency graph is established, DataIDS monitors the cross-correlation indices between connected sensors over time. A change-detection index ($R_i(t)$) is calculated for each sensor, summing the cross-correlations with its neighbors in the graph. A detection occurs when this index drops below an automatically computed threshold ($\Theta$).
  • Attack Isolation: Upon detecting a potential attack, an isolation procedure is initiated. This involves iteratively removing sensors from the set of potentially compromised devices and re-evaluating the cross-correlation indices. If removing a sensor causes the cross-correlation index to return to normal, that sensor is identified as the attacker.
  • Attack Tree Modeling: The paper also utilizes an attack tree to model potential threats and vulnerabilities in Fog/IoT scenarios, helping to assess risks and guide countermeasure selection.

Datasets

The paper describes a test bed setup at the University of Florence for experimental validation. This test bed consists of 12 DHT11-DHT22 devices generating humidity and temperature data streams. The dataset comprises approximately 10,430 samples acquired every 5 minutes. This dataset was used to simulate and evaluate the performance of DataIDS against various attack models. No new benchmark datasets were introduced; rather, a real-world experimental setup was utilized.

Key Insights and Ideas

  • Data-Centric Security is Crucial for IoT: The paper strongly emphasizes that traditional network-level security is insufficient for IoT. Attacks targeting the actual sensor data can be subtle and require a different detection paradigm.
  • Fog Computing as a Security Enabler: By bringing computation closer to the edge, Fog Computing significantly enhances the feasibility and efficiency of real-time data analysis for security purposes, overcoming the limitations of resource-constrained IoT devices.
  • Dependency Graphs Capture Heterogeneous Relationships: The novel use of dependency graphs, based on cross-correlation, effectively models relationships between diverse sensor types. This is a key insight for understanding system behavior and detecting anomalies that might not be apparent through other means.
  • Attack Isolation is Achievable: The proposed iterative isolation procedure demonstrates a practical method to pinpoint the source of false data injection, which is critical for effective remediation.
  • Attack Tree for Risk-Informed Countermeasures: The integration of attack tree analysis provides a structured way to match countermeasures to the specific attack and its potential impact, avoiding disproportionate security measures.

Key Results

The experimental results demonstrate the effectiveness of DataIDS:

  • High Detection Rate: Cyber attacks were detected in 97.5% of the cases across various attack types.
  • No False Positives: The system achieved a 100% precision rate, meaning no false positive detections were observed.
  • Good Isolation Capability: The isolation capability worked in 82.5% of the cases, correctly identifying the attacked sensor.
  • Attack Type Sensitivity: Replay, stuck-at, and dynamic perturbation attacks were detected and isolated with very high accuracy (100% detection, 96.7% isolation). The noise addition attack proved more challenging for isolation (90% detection, 40% isolation).
  • Prompt Detection: The detection times (min, MW, and max) indicate that the system can promptly identify attacks.
  • Accurate Time Estimation: The estimation of the attack start time ($f$) was shown to be excellent, allowing for the discarding of erroneous data.
  • Computational Efficiency: While dependency graph creation takes around 36.2 seconds, detection and isolation phases are very fast (26.4 ms per unit), making it suitable for deployment on resource-constrained devices.

Limitations and Future Work

  • Single Attack Assumption: The isolation procedure implicitly assumes that only one IoT device in a cluster is attacked. If multiple sensors are under attack simultaneously, the assumption may be weakened.
  • Attack vs. Fault Distinction: The approach cannot inherently distinguish between attacks and sensor faults. However, the authors suggest that rapid detection and countermeasure implementation can help infer a sensor fault.
  • Future Work:
    • Data Privacy: Extending the model to better address data privacy concerns in Fog/IoT environments.
  • Cooperative Intrusion Detection: Enabling cooperative intrusion detection capabilities among logically separated sensor zones.
  • Dependency Graph Properties: Investigating how dependency graph properties (size, number of connected vertices) influence attack detection and isolation capabilities.

Summary updated

Abstract

The Internet of Things (IoT) context brings new security issues due to billions of smart end-devices both interconnected in wireless networks and connected to the Internet by using different technologies. In this paper, we propose an attack-detection method, named Data Intrusion Detection System (DataIDS), based on real-time data analysis. As end devices are mainly resource constrained, Fog Computing (FC) is introduced to implement the DataIDS. FC increases storage, computation capabilities, and processing capabilities, allowing it to detect promptly an attack with respect to security solutions on the Cloud. This paper also considers an attack tree to model threats and vulnerabilities of Fog/IoT scenarios with heterogeneous devices and suggests countermeasure costs. We verify the performance of the proposed DataIDS, implementing a testbed with several devices that measure different physical quantities and by using standard data-gathering protocols.
BibTeX
@inproceedings{fantacci2019false,
  author = {Romano Fantacci and Francesca Nizzi and Tommaso Pecorella and Laura Pierucci and Manuel Roveri},
  title = {{False Data Detection for Fog and Internet of Things Networks}},
  booktitle = {Sensors},
  year = {2019},
  url = {https://openreview.net/forum?id=9EyzlmGc4Rc},
}
Figure