- Designed a convolutional neural network to identify non-recurring traffic congestion, achieving a 98.73% accuracy with low false positive and false negative rates.
- Developed a short-term bus delay prediction model that combines unsupervised clustering analysis and Kalman filters. The root-mean-square-error is only 60 seconds, which outperforms the state-of-the-art in accuracy.
- Applied long-term predictive analytics on historical General Transit Feed Specification (GTFS) and time-point bus data through MongoDB using scikit-learn and Matplotlib Python libraries.
- Developed unsupervised mechanisms for optimizing the on-time performance of fixed schedule transit vehicles.
- Designed and implemented the T-HUB iOS App which features route planning, delay estimation and real-time navigation, using Objective-C, Core Data, Google Map SDK for iOS, GTFS, GPS and RESTful APIs, and it’s used by hundreds of bus riders in Nashville.
Transit Hub - Predicting and Optimizing the Performance of Urban Mobility
(Funded by the National Science Foundation (NSF) and the Siemens Corp.)
Robust Software Modeling Tool - Detecting Cyber-Attacks Using Machine Learning and Unit Tests
(Funded by the Office of Naval Research (ONR))