Because sensor networks are often deployed in hostile environments where their security and integrity may be
compromised, it is essential to maximize the reliability and trustworthiness of existing and envisioned sensor networks.
During operations, the sensor network must be robust to deception, node compromise, and various other attacks, while
maintaining the operator's situational awareness regarding the health and integrity of the system. To address these needs,
we have designed a Framework to Ensure and Assess Trustworthiness in Sensor systems (FEATS) to identify attacks on
sensor system integrity and inform the operator of sensor data trustworthiness. We have developed and validated
unsupervised anomaly detection algorithms for sensor data captured from an experimental acoustic sensor platform
under a number of attack scenarios. The platform, which contains four audio microphones, was exposed to two physical
attacks (audio filtering and audio playback) as well as a live replay attack (replaying live audio data that is captured at a
remote location), which is analogous to a wormhole attack in the routing layer. With our unsupervised learning
algorithms, we were able to successfully identify the presence of various attacks.