Inter-District Mode Choice Modelling of Resident Students of BUET
I studied how BUET resident students choose between bus, AC bus, train, personal car, and rental for inter-district trips from Dhaka. I designed an RP questionnaire, collected and cleaned data from 241 students, and estimated discrete choice models in R with Apollo. I built thirteen models, ranging from simple to complex specifications, and then selected the final model using log likelihood, AIC, BIC, adjusted R-squared, and t-tests. Full description can be found here.
DhakaSim: Making Crash Detection Realistic for Laneless, Mixed Traffic
DhakaSim is a microsimulation of dense, laneless urban corridors with mixed users (cars, rickshaws, bikes, pedestrians). The original build over-reported “accidents” in queues and during merges because close passes and mid-merge overlaps were counted as crashes. I redesigned the safety logic to (i) distinguish low-severity contacts from true crashes using 2D body overlap and closing-speed thresholds, (ii) evaluate lane changes with multi-step (swept) safety checks that abort unsafe merges, and (iii) handle pedestrian/object interactions deterministically via geometry + speed. Parameters were exposed for calibration, and structured logs now separate contacts, minor crashes, and major crashes. Across paired scenarios, reported accidents dropped substantially (e.g., mean 36.0 → 14.9 per run; ~59% overall reduction) while preserving detection of higher-severity events—supporting credible, severity-aware KPIs and fair evaluation of low-cost treatments. Full description can be found here.
Adaptive Traffic Signal Control with SUMO and Python
I designed and implemented a simulation-based study to assess the impact of adaptive traffic signal control on urban mobility. Using SUMO (Simulation of Urban Mobility) and Python’s TraCI API, I built a small grid network in NetEdit with multiple intersections and realistic vehicle flows. The project compared fixed-time signals against a queue-responsive adaptive controller written in Python. The adaptive controller dynamically adjusted green times based on real-time queue lengths, while ensuring minimum and maximum green durations for safety and stability. Full description can be found here.
Metro smart locker adoption (Dhaka Metro Rail Line 6)
I am building an equity-aware behavioral framework to understand who adopts station smart lockers and why. I combine latent segmentation with structural modeling to capture heterogeneity in preferences, then test how design choices like fee levels, locker location relative to entrances, and friction in payment and authentication change adoption intent. This supports evidence-based decisions on where to deploy lockers, how to price them, and which user groups may need targeted design adjustments.