Using Artificial Intelligence to Improve Traffic flows, with Consideration of Data Privacy

This project develops an artificial neural network (ANN), a class of Artificial Intelligence (AI) systems, to accurately model and predict future delays at an intersection. Developing such modeling and prediction systems raises considerable data privacy concerns and it is incumbent upon municipal, state, and federal branches of government to prioritize citizens and their concerns before the implementation of new smart community technologies that are fueled by unprecedented levels of data collection. The technique proposed in this study identifies nonlinear, time-varying mapping between the inputs to the ANN and its output, the predicted delay. The traffic data measured at a Long Beach intersection with heavy truck traffic are used to build a realistic simulation in Vissim, a microscopic traffic flow simulator. The authors designed and performed experiments on the developed Vissim model to train the ANN delay predictor and validate the generalization ability of the predictor. The simulation results agree with the on-site delay measurements. This suggests the ANN predictor can accurately predict the delay at the intersection with heavy-truck penetration. Because smart technologies raise data privacy concerns, the research team led 32 study participants on “datawalks” designed to gauge comfort levels and attitudes toward devices that collect personally identifiable information. Study participants encountered public WiFi routers, surveillance cameras, automated license plate readers and other surveillance technologies. They used a custom app to respond to prompts related to data collection, sharing and analysis. Study participants’ responses, along qualitative data collected during a “debriefing” conversation following each walk, provided insights into residents’ attitudes toward smart communities technologies and identified privacy concerns. The quantitative and qualitative findings in this study inform a series of recommendations that research teams can follow to implement real-world test labs at busy truck intersections while fostering public trust, installing these modelling and prediction systems, and ensuring the overall safety and efficiency of the intersection’s traffic flow. The full report is available from the US Department of Transportation's National Transportation Library at https://rosap.ntl.bts.gov/view/dot/67414