Ali Arman — Behavioral Modeling

Behavioral Modeling

Behavioral modeling and analysis focus on understanding, quantifying, and predicting how individuals and groups make decisions under constraints, uncertainty, and trade-offs. At its core, it treats behavior as an observable outcome of latent preferences, perceptions, incentives, and contextual factors, translated into measurable choices and actions. This perspective is widely applicable across domains, ranging from mobility and energy use to marketing, finance, and human–AI interaction, because it combines rigorous statistical modeling with data-driven inference to explain why decisions are made, not only what happens. For data science practice, behavioral analysis provides a structured way to link raw data to interpretable insights, policy levers, and operational decisions.

Travel behavior analysis applies this behavioral lens to how people organize activities and choose when, where, how, and whether to travel. Its primary purpose is to explain and forecast demand by explicitly modeling individual decision-making rather than relying solely on aggregate trends. Applications include travel demand forecasting, pricing and incentive design, mode choice and accessibility analysis, evaluation of new mobility services, and assessment of policy interventions related to sustainability and equity. By connecting socio-demographics, activity patterns, network conditions, and perceived costs, travel behavior analysis supports evidence-based planning and business decisions in contexts where demand is elastic, heterogeneous, and sensitive to both economic and non-economic factors.

Driving behavior analysis shifts the focus from trip-level decisions to the tactical and operational choices made during driving, such as speed selection, car-following, gap acceptance, lane changing, and responses to traffic control or assistance systems. Its objectives include understanding heterogeneity in driving styles, identifying precursors to safety-critical events, and quantifying how drivers adapt to infrastructure, traffic conditions, and vehicle technologies. Practical applications span road safety assessment, risk modeling, advanced driver-assistance systems (ADAS), connected and automated mobility, insurance analytics, and real-time traffic operations. From a data science perspective, driving behavior analysis bridges human factors with high-resolution sensor data, enabling interpretable models that inform both system design and real-time decision support.

The methodological foundation of behavioral analysis combines econometric modeling with experimental and data-centric approaches. Discrete choice analysis models decisions as outcomes of utility maximization, allowing preferences, trade-offs, and heterogeneity to be estimated in a statistically rigorous and interpretable manner. Design of experiments provides principled strategies to generate informative data by systematically varying attributes and conditions, improving identifiability and efficiency. Revealed preference methods infer behavior from observed real-world choices, capturing actual constraints and habits, while stated preference methods elicit responses to hypothetical scenarios to evaluate new policies, products, or technologies before deployment. Together, these methods form a coherent toolkit that is transferable across domains wherever understanding and predicting human decision-making is essential.

Across my projects, I have systematically applied the full spectrum of behavioral analysis methods: discrete choice analysis, rigorous data collection survey and companions for revealed preference analysis, and design of experiments for stated preference studies, covering the entire pipeline from hypothesis and assumptions definition to data collection to modeling, inference, and decision support. More recently, I have expanded this methodological foundation with unsupervised, neural-network-based classification and representation learning to uncover latent behavioral patterns and address behavioral heterogeneity at scale. This combination allows me to move seamlessly between interpretable, theory-driven models and data-driven discovery, enabling robust segmentation, explanatory insight, and predictive performance in complex, real-world settings where behavior is diverse, dynamic, and only partially observable.

Unsupervised Profiling of Tactical Driving Styles from Large-Scale Trajectory Data

This project uncovers tactical driving behavior directly from large-scale panel floating car trajectory data collected along an extended motorway corridor. Rather than relying on predefined behavioral labels, the objective was to reveal naturally emerging driving patterns, such as aggressiveness, conservativeness, and adaptation to congestion, through data-driven analysis of longitudinal and lateral maneuvers. The motivation stems from a critical gap in traffic safety and operations: most traffic management and risk assessment models assume homogeneous driver behavior, while real-world traffic is inherently heterogeneous. By extracting latent tactical styles from real trajectories, the study provides a behaviorally grounded basis for safety assessment, congestion management, and connected and automated mobility strategies.

Methodologically, the study integrates high-resolution trajectory reconstruction with feature engineering of maneuver-level indicators and advanced unsupervised learning techniques. Dimensionality reduction and clustering frameworks were applied to characterize multi-dimensional behavioral signatures, capturing speed adaptation, lane-change dynamics, and spacing strategies under varying traffic states. The approach rigorously separates tactical behavior from purely contextual traffic effects, enabling interpretable segmentation of drivers across time and space. The combination of large-scale panel data, spatiotemporal feature extraction, and unsupervised modeling demonstrates the ability to handle noisy, high-dimensional datasets while preserving behavioral interpretability, reflecting advanced competence in statistical learning, representation modeling, and traffic flow theory (Transportation Research Symposium - Transportation Research Procedia - 2025).

Beyond transportation, this work exemplifies transferable data science capabilities in behavioral segmentation, anomaly detection, and pattern discovery within complex dynamic systems. The pipeline, data fusion, scalable preprocessing, feature construction, unsupervised classification, and interpretability analysis, mirrors industrial applications such as customer segmentation, fraud detection, insurance risk profiling, telematics analytics, and predictive maintenance. Identifying latent behavioral archetypes from raw time-series data enables targeted intervention, personalized services, and risk-aware decision support. The project highlights the ability to convert massive unstructured spatiotemporal data into structured, economically meaningful behavioral intelligence applicable across domains where heterogeneity and dynamics drive performance and risk.

Uniqueness

Large-scale unsupervised behavioral segmentation from high-resolution motorway trajectory panel data.
Identification of latent tactical driving styles for proactive safety and risk assessment.
Data-driven support for C-ITS, ADAS calibration, and dynamic traffic management strategies.
Transferable framework for time-series clustering, anomaly detection, and behavioral profiling in industrial analytics.

Modeling Preferences for Stop-Based Pooled Automated Mobility

This project investigates how travelers perceive and evaluate an on-demand, stop-based pooled automated vehicle (AV) service within a real-world multimodal transport system in Flanders, Belgium. The core objective was to understand whether such a service would substitute private cars, compete with public transport, or complement existing mobility options, while explicitly accounting for differences across trip purposes (work/school, leisure, and groceries). The study addresses a critical policy and industry question: how can shared automated mobility be integrated sustainably without undermining public transport or inducing additional car dependence? The findings provide actionable insights for regulators, operators, and technology developers designing scalable automated mobility services.

Methodologically, the project combines a large-scale stated choice experiment (652 respondents, nearly 5,000 binary observations) with advanced discrete choice modeling, including Multinomial Logit (MNL), Nested Logit (NL), and Cross-Nested Logit (CNL) formulations. The use of cross-nested structures enabled rigorous identification of substitution patterns and latent similarity between the proposed AV service and conventional modes. The modeling framework formally derived and estimated correlation structures in unobserved utility components, quantified substitution strength, and estimated value of travel time (VTT) across trip purposes. This level of structural modeling, combined with experimental design grounded in revealed reference trips, demonstrates the ability to move beyond surface-level prediction toward behaviorally interpretable, statistically robust, and policy-relevant inference (Transportation Research Board 103rd Annual Meeting - 2024 & European Transport Research Review - 2025).

Beyond transportation and mobility, the project showcases transferable skills highly relevant to industrial data science and economic analytics. It integrates experimental design, advanced econometrics, probabilistic modeling, heterogeneity analysis, and large-scale optimization-based estimation (via Python Biogeme and a tailored weighted maximum likelihood estimator). The ability to quantify trade-offs (time vs. cost), identify latent substitution structures, estimate willingness-to-pay, and model behavioral segmentation under uncertainty is directly applicable to pricing strategy, product design, market-entry analysis, demand forecasting, and policy evaluation in sectors such as energy, insurance, logistics, and digital platforms. The work reflects a capacity to translate complex behavioral data into structured decision-support models with clear economic interpretation and measurable strategic value.

Uniqueness

Advanced cross-nested logit modeling to uncover latent substitution patterns and correlation structures across travel modes.
Evidence-based positioning of pooled AVs as a hybrid between bus and car, guiding sustainable mobility strategy.
Quantified value-of-travel-time differentials across trip purposes for pricing, revenue, and policy design.
Behavioral segmentation insights enabling targeted deployment strategies across income groups and trip contexts.