Enhancing Data Privacy in Edge-based Driver AI Monitoring Systems Through Adaptive Differential Privacy

Edge-based computing has appeared as a powerful paradigm to minimize latency and dependency on network delivery of data, with many intelligent systems now allowing for data to be processed locally, thus providing responses more quickly to the user. For instance, in driver monitoring systems (DMS), the privacy of sensitive driver data is handled in many edge-based scenarios which raises privacy and security concerns due to these non-traditional decomposition. Implicitly, current privacy-preserving techniques: data-disaggregation, anonymization, homomorphic encryption, etc. can get it wrong in certain analyses – such as driving behaviour analyses where inference is real-time. The research responds to this problem of data privacy without compromising on analysis accuracy in edge-based DMS. Hence, it examined whether it could introduce differential privacy using the Laplace and Gaussian noise methods which would help to preserve personal information but allow the important identification of driver behaviour, e.g. fatigue, distraction, unsafe actions, etc. We propose a method of differential privacy which is simply an adaptive differential privacy method that provides dynamic noise based on the context and sensitive environment of one's data. The results are novel, and provided a remarkably 28.7 % reduction in privacy leakage, +1.7% in utility (accuracy) and 10-20 % shorter latencies than existing differential privacy methods. Overall, this system premise presents a very balanced approach between privacy/accuracy/timeliness that follows to within the range of privacy-aware, real-time behavioural analytics with far better prospects in an edge-based computing space.