Source-linked AI summary
FIRST: A Framework for Optimizing Information Quality in Mobile Crowdsensing Systems
Francesco Restuccia, Pierluca Ferraro, Timothy S. Sanders, Simone Silvestri, Sajal K. Das, Giuseppe Lo Re
TL;DR
Mobile crowdsensing needs reliable classification despite malicious or unreliable participant reports. FIRST uses mobile trusted participants and optimization algorithms to determine their minimum deployment before system deployment, achieving about 80% classification accuracy in the reported evaluations.
Problem
Mobile crowdsensing lacks reliable participant behavior and requires accurate classification of reliable versus unreliable sensing reports.
Method
FIRST combines trusted participants with CVP, LEA, and MOA to estimate report validation and optimize the minimum MTP deployment for target accuracy.
Results
About 80% average classification accuracy, an improvement of about 20%, was achieved in the practical evaluation.
Takeaways & Limitations
FIRST reduced the impact of corruption, on/off, and collusion attacks in the considered scenarios.
Takeaways & Limitations
LEA relies on strong assumptions and provides an approximate, rather than fine-grained, mobility estimate before deployment.
Abstract
from arXiv · showhide
Mobile crowdsensing allows data collection at a scale and pace that was once impossible. One of the biggest challenges in mobile crowdsensing is that participants may exhibit malicious or unreliable behavior. Therefore, it becomes imperative to design algorithms to accurately classify between reliable and unreliable sensing reports. To this end, we propose a novel Framework for optimizing Information Reliability in Smartphone-based participaTory sensing (FIRST), that leverages mobile trusted participants (MTPs) to securely assess the reliability of sensing reports. FIRST models and solves the challenging problem of determining before deployment the minimum number of MTPs to be used in order to achieve desired classification accuracy. We extensively evaluate FIRST through an implementation in iOS and Android of a room occupancy monitoring system, and through simulations with real-world mobility traces. Experimental results demonstrate that FIRST reduces significantly the impact of three security attacks (i.e., corruption, on/off, and collusion), by achieving a classification accuracy of almost 80% in the considered scenarios. Finally, we discuss our ongoing research efforts to test the performance of FIRST as part of the National Map Corps project.
1 INTRODUCTION
Mobile crowdsensing expands data collection through pervasive smartphones but depends on reliable participant reports. FIRST addresses malicious or unreliable behavior by optimizing trusted-participant deployment and achieves about 80% classification accuracy in its evaluations.
- Smartphones with embedded sensors and wireless connectivity enable mobile crowdsensing at a scale beyond static sensor deployments.
- Motivations and Challenges: Malicious or unreliable participants can submit false reports, threatening information quality and application reliability.
- Motivations and Challenges: MTPs securely bootstrap trust by submitting reliable reports that help build participant reputation scores, but they add system cost.
- Contributions: FIRST formulates the MTP Optimization Problem to minimize deployed MTPs while guaranteeing desired classification accuracy.
- Contributions: FIRST combines CVP, LEA, and MOA to estimate validation likelihood, approximate sector likelihood from geography, and solve the deployment optimization problem.
- Contributions: About 80% average classification accuracy, improving by about 20%, was achieved in the practical evaluation.The evaluation used an iOS and Android crowdsensing app distributed to volunteers at IEEE PerCom 2015.
2 RELATED WORK
Related work improves information reliability mainly through trusted hardware or reputation-based systems. FIRST differs from participant-recruitment approaches by using mobile trusted participants to compute trustworthiness and improve information quality.
- Existing information-reliability approaches include trusted platform modules and reputation-based systems.
- Trusted platform modules: Trusted platform modules authenticate sensor data but require additional hardware unavailable on ordinary off-the-shelf devices.
- Reputation-based systems: Reputation-based systems estimate and update user reputation over time to improve sensing-report reliability.
- Participant recruitment: Participant-recruitment frameworks target sensing coverage, localization, energy efficiency, or incentive costs rather than trustworthiness.
- Participant recruitment: FIRST instead relies on MTPs to compute participant trustworthiness, ultimately improving information quality significantly.
3 PRELIMINARIES AND BACKGROUND
The system models sensing reports, participant behavior, and trusted-participant validation in a sectorized mobile crowdsensing architecture. It defines classification accuracy and an optimization problem for selecting trusted participants under deployment constraints.
- The MCS architecture connects participants and the platform through mobile apps, cloud-delivered sensing requests, and Internet access.
- The monitored phenomenon is assumed quantifiable, dynamic, and not subject to personal opinion.
- The model focuses on malicious and unreliable participant behavior while assuming secure platform functions and no Sybil or rejoin attacks.
- Mobile Trusted Participants: MTP reports remain valid for a parameterized time window T whose suitable duration depends on sensing variability.The paper gives 5–10 minutes as an example for traffic monitoring and a longer duration for gas-price monitoring.
- Mobile Trusted Participants: Reports are validated when a recent MTP report exists in the same sector; matching reports are classified reliable, while nonmatching reports are rejected.
- MTP Optimization Problem: The MTP Optimization Problem addresses limited recruitment budgets because more MTPs increase validated reports and information reliability.
- MTP Optimization Problem: Classification accuracy is defined as 1 − P{E}, where P{E} is the probability of erroneous classification.
4 THE FIRST FRAMEWORK
FIRST estimates sensing-report validation from participant and MTP mobility, then uses this estimate to optimize the minimum MTP deployment needed for reliable classification. Its components combine probabilistic validation modeling, map-based mobility approximation, and efficient optimization, with extensions for trust inference and attack resilience.
- Computation of Validation Probability: FIRST computes the probability that a sensing report is validated by at least one MTP as a function of MTP count and mobility.The CVP component derives this validation probability for users with potentially different mobility distributions.
- Computation of Validation Probability: With five MTPs and eight sectors, the example gives P{V}=0.49 for S1 and P{V}=0.71 for S2.S1 assumes uniform mobility, whereas S2 assigns greater validation likelihood to sectors near likely MTP locations.
- Likelihood Estimation Algorithm: LEA uses map information, popular places, and arterial roads to approximate user and MTP mobility when prior mobility data are unavailable.The method divides the sensing area into sectors, extracts geographical constraints, and processes a map image to estimate sector likelihoods.
- Likelihood Estimation Algorithm: LEA approximates mobility distributions using real-world traces from Rome, San Francisco, and Beijing, but relies on strong assumptions and can be complemented by better post-deployment mobility information.The authors describe LEA as a pre-deployment heuristic rather than a fine-grained mobility estimator.
- Solving the MTP Optimization Problem: FIRST uses P{R | V } to classify non-validated reports by inferring user trustworthiness from previously validated MTP-supported reports.The framework applies Jøsang’s trust model to estimate whether a user’s future reports are reliable.
- Solving the MTP Optimization Problem: MOA uses a modified binary search to find the minimum MTP count satisfying the target error bound, with LEA reducing its complexity to O(n · logmmax).The complexity reduction follows from assuming users and MTPs share the same mobility distribution under LEA.
5 EXPERIMENTAL RESULTS
Experiments with mobility traces and a participatory attendance system evaluate FIRST’s components, resource requirements, and resilience to attacks. FIRST generally outperforms comparison methods, achieving high accuracy with controlled MTP deployment.
- Experimental setup: FIRST was evaluated on vehicular traffic mobility traces and a practical participatory attendance implementation against FIDES,, and majority vote.The evaluation considered corruption, on/off, and collusion attacks.
- Evaluation of FIRST components: 3.47% maximum difference shows that CVP-LEA closely estimates validation probability across the three mobility scenarios.CVP-Uniform reached a 17.02% maximum difference in Rome.
- Evaluation of FIRST components: 85% average accuracy required approximately 32% MTP density in Rome and Beijing, versus 55% in San Francisco.The results used the MTP Optimization Algorithm across unreliable-report probabilities P{F} = 0.01, 0.5, and 0.9.
- Evaluation of FIRST components: P{F} = 0.5 is the worst case for MTP requirements because participants are maximally unpredictable, while San Francisco requires the most MTPs under comparable targets.The mobility setting and participant behavior both affect the required MTP density.
- Evaluation of attack resiliency: Below 7% error rate, FIRST tolerated collusion with 400 MTPs regardless of attacker or collusion-group count.The authors attribute this resilience to MTP validation rather than data aggregation.
- Participatory PerCom: 76.02% average accuracy exceeded FIDES at 64.45%, at 63.99%, and majority vote at 46.2% in the attendance experiment.FIRST remained superior even when MTPs reported every 40 minutes and only 25% of reports were verified on average.
6 CONCLUSIONS AND FUTURE WORK
The paper concludes that FIRST models and optimizes information reliability using MTPs, LEA, and MOA, and that evaluations show advantages over existing approaches. Future work applies the framework to the National Map Corps context.
- Conclusions: FIRST models and optimizes information reliability through MTPs, the likelihood estimation algorithm LEA, and the MTP optimization algorithm MOA.The framework was evaluated with real mobility traces and a practical conference-attendance system.
- Conclusions: Evaluations found that FIRST outperforms existing work in increasing information reliability and accurately capturing system performance.The conclusion summarizes results from traffic sensing and participatory attendance experiments.
- Future work: The ongoing National Map Corps collaboration is intended to test FIRST and other mobile crowdsensing systems at larger scale.The prototype supports updating, deleting, and retrieving points of interest in real time.