Nicholas Fournier, PhD

Development and implementation of an open source activity-based travel model ActivitySim. Notable implementations for San Diego and Dubai. Created a visitor model and developed a disaggregated accessibility measure estimator, linking higher level models to lower level choice models.

Design of data processing pipeline to fuse form-based and smartphone-based travel survey data, impute missing values, adjust for bias, and reweighted to the target region's population using PopulationSim. Pipeline includes Postgres, R, Python, and visualization in Rmarkdown generated HTML flex dashboard.

Analytical simulation exploring revenue and traffic flow with “futures” market toll pricing. Utilized Kernel Density Estimation to smooth traffic flow data for forecasting and pricing models. Funded by California State SB1.

Ongoing research to develop generalized bicycle network connectivity performance measure using graph theory, open-data, and user preference criteria (e.g., route choice models). Funded by Caltrans.

Python program to detect operational but mislabeled traffic sensors using variety of machine learning techniques (e.g., k-Nearest Neighbor, Logistic Regression, Random Forest, Support Vector Machines, Local Outlier Factor, Isolation Forest, and Robust Covariance Anomaly Detection). Funded by Caltrans.

UX research/human factors study to determine bicycle infrastructure preferences using virtual reality bicycle simulator. Estimated using a Latent Class Choice Model capable of accounting for user heterogeneity. Results to align with “Complete Cities” project. Funded by Caltrans.

Advised team of 19 doctoral students conducting industry partnered research in public transportation engineering, planning, policy analysis, and economics.

Data fusion population synthesizer using novel combinatorial optimization algorithm in R and C++ (demographics, OD-matrices, household/vehicle association, etc.). Mixed-methods include Bayesian Networks, Markov chain Monte Carlo simulation, iterative fitting (matrix raking), robust regression, LASSO/Ridge regularization, and gradient descent. Used as input in larger agent-based discrete choice and simulation to lower energy consumption with user incentives. Joint MIT project sponsored by ARPA-energy.

Utilized a driving simulator to test driver response to novel infrastructure treatments, such as bicycle infrastructure and dynamic signage for the visually impaired. The driving simulator is a full sized vehicle with its engine removed, fitted with sensors to all user inputs and responses (eye tracking, pedals, steering wheel, shifter, radio, etc.) and surrounded by projector screens for an immersive user experience. Sponsored by the U.S. DOT SaferSim UTC.