Real-Time Threat Assessment with Imperfect Sensor Data is an open-access Naval Postgraduate School thesis by Fredrick B. Stanford.
The Department of Defense employs various sensors daily to gather crucial data about potential threats to the United States and its allies. However, these sensors do not always capture the ground truth due to technological limitations and human errors. This thesis introduces a mathematical framework for utilizing sensor-collected data that may contain false negatives and false positives to detect threats effectively.
The first formulation assumes that the sensor operator has intelligence regarding the likelihood of threats at specific locations, while the second formulation considers that adversaries may deliberately choose targets to avoid detection. In both cases, the author develops a threshold-based policy that directs the sensor to areas where an attack is most likely occurring, triggering an alarm if the probability of an attack exceeds a location-specific threshold.
The author employs Monte Carlo simulations to evaluate these threshold-based policies against two conflicting objectives: maximizing the probability of real-time threat detection and minimizing the average time between false alarms. The findings of this research enable our forces to quantify imperfect sensor data using robust algorithms, moving beyond ad-hoc assessments and relying on the expertise of subject matter experts.
Real-Time Threat Assessment with Imperfect Sensor Data contains the following major sections:
- Introduction
- Sensor Operation with Intelligence on Attack Likelihood
- Sensor Operation Against Strategic Attackers
- Conclusion
Post Image Credit- Adobe Stock by Deenanath
