Back to Projects
Data Science & Mobility Intelligence • 2025The Vision & Architecture
Decoding urban transit dynamics with
rigorous statistical distributions.
Flux is an exploratory data analysis and inferential statistics pipeline processing over 3.6 million passenger transaction records from the Shanghai Metro merged with localized weather data. Proves Poisson arrival failure, fits Negative Binomial overdispersion, and validates the Central Limit Theorem via Monte Carlo simulations.
Core Capabilities
Overdispersion Testing & Distribution Fitting
Calculated a dispersion ratio of ~388.7 (variance/mean), formally rejecting the Poisson process via Kolmogorov-Smirnov test and fitting a Negative Binomial model to prevent underestimating peak overcrowding.
Empirical Central Limit Theorem Simulation
Designed Monte Carlo sampling simulations (n=100) demonstrating convergence of highly skewed passenger distributions into Gaussian normal curves (Jarque-Bera p > 0.05).
Weather Impact Sensitivity Modeling
Quantified meteorological effects, proving rain heavily depresses discretionary leisure travel (NHBinFlow) while non-discretionary commuter flow (CinFlow) remains rigid.
Spatiotemporal Mobility Signatures
Identified distinct weekday dual-peak "M" commuter profiles (7–9 AM, 5–7 PM) versus weekend single-peak bell curves across 16 metro lines and critical transfer hubs.
Operational Strategy & Policy Engineering
Translated statistical confidence intervals into transit policies, including gap-train insertions at residential bottlenecks and weather-contingent staff redeployment.
Technology Stack
Architectural Decisions
Negative Binomial over Poisson distribution
The extreme overdispersion (variance 400x mean) causes standard Poisson models to drastically underestimate peak crowd risks.
In-memory vectorized Python over distributed Spark
Processing 3.6 million rows in-memory via Pandas avoids cluster serialization overhead and executes statistical tests in seconds.