In a wide range of settings, various interconnected devices and sensors are used to collect data (e.g., cameras, piezoelectric, acoustic, environmental, and magnetic sensors). Data obtained from these sensors are used to support various types of services depending on the objective of the sensor network. The accuracy and integrity of collected data are crucial to the reliability of such services and the optimal performance of the network. Recent research has shown that hackers could compromise the sensors and send misleading data to the controller, potentially causing severe disruptions. For example, in a smart-city setting where the sensors are used to monitor traffic patterns, hackers could cause significant traffic problems and compromise the entire operation of the smart city services.

This project breaks the problem of integrity assurance in an Internet of Things (IoT) network into a two-stage process. The first stage focuses on the detection and identification of anomalous data. The second stage focuses on decision support in the presence of anomalous data, i.e. what to do once untrustworthy data has been detected and how to adjust the decision-making process in the presence of untrustworthy data. This project is a joint effort by ULL and UVA.

ULL team has focused on spatial analysis. A spatiotemporal model was developed to analyze the neighboring sensors and study various characteristics of the traffic data such as the peak hour, and the volume of the traffic in each direction. The analysis was conducted on real-time traffic data sourced from the Pennsylvania Department of Transportation. Spatial-temporal Kriging models were used to establish the predictability of traffic flow from different locations in the study area. Four different covariance models were fitted to the targeted data and their prediction performance was evaluated based on the root mean squared error (RMSE). The models showed high performance with RMSE of 0.394. Due to memory limitation in R, a comparison between all four models is left for future extension.

UVA focus was on trust models and decision support. To accomplish optimal decision support, the trust score for individual datum must be accurate. Although originally tasked with developing decision support in the presence of untrustworthy data, the UVA effort shifted to primarily focus on developing new trust scoring models. Two methods were developed. The first was an extension of the model developed during Year 6. The second was a new ensemble method for calculating trust scores that utilizes numerous models built using data from the other nodes in the IoT network. The decision support development was left to future work.

While IoT is increasingly adopted in many domains including critical infrastructures, very little work has discussed the best approaches to identify anomalies pertaining to cybersecurity and stealthy data injection attacks. From a detection point of view, the design of our model is unique in that it guarantees and provides data integrity validation from both temporal and spatial perspectives. Adaptive Kalman filtering is also used specifically for traffic prediction, but largely depends only on historical data due to limitations in our data set. The outcomes of the project will be generally applicable for any real-time data gathering in IoT deployments, but specifically provides highly accurate results for intelligent traffic management due to availability of training data sets. Current research on the trustworthiness of data focuses on detection and often overlooks the decision-making problem. The proposed project is unique in the fact that it will develop a decision support tool that incorporates an integrity or trust score into the decision-making process.

Outcomes of this project will directly impact the reliability, trustworthiness, resiliency, and integrity of a wide range of real-time data collection in IoT environment including smart city services, emergency and disaster managements, urban sensing and law enforcements. Companies serving the public sector (i.e. government agencies) can take advantage of these techniques to improve the security and reliability of IoT sensor data. This will also directly support many CGI and (e.g., Smart City Air Challenge). Results will be also of great interest to companies providing real-time ubiquitous services such as the healthcare domain.