Kamal Acharya

Postdoctoral Research Associate at Baylor University

Back to Publications

Doctoral Dissertation

Enhancing Demand Modeling for Advanced Air Mobility Using Data-Driven Learning and Neurosymbolic AI

By Kamal Acharya

A behaviorally grounded, data-driven, and interpretable framework for forecasting demand in emerging Urban and Regional Air Mobility systems.

2026 University of Maryland, Baltimore County (UMBC) Doctor of Philosophy in Information Systems

Advisor: Dr. Houbing Song Co-advisor: Dr. Alvaro Velasquez

Advanced Air Mobility Demand Modeling Neurosymbolic AI

Download Dissertation Cite

Abstract

Advanced Air Mobility (AAM), including Urban Air Mobility (UAM) and Regional Air Mobility (RAM), faces fundamental challenges in demand modeling due to the absence of historical operational data, strong spatial and temporal heterogeneity in traveler behavior, and the dominance of black-box machine learning approaches that lack interpretability and behavioral grounding. These limitations restrict the reliability of demand forecasts for policy evaluation, infrastructure planning, and operational design. This dissertation addresses three interconnected challenges in AAM demand modeling: behaviorally grounded adoption modeling, structurally constrained data-driven demand estimation, and interpretable learning under uncertainty. First, a Generalized Cost of Trip framework models AAM adoption by integrating monetary cost, travel-time valuation, reliability, and perceived risk across regional Tennessee and high-resolution New York City case studies. Second, learning-based methods are integrated with classical optimization and spatial interaction structures through a neural network–accelerated genetic algorithm and a machine-learning-enhanced gravity model. Third, a neurosymbolic framework embeds decision-tree-derived symbolic rules within neural networks to jointly support predictive accuracy, robustness under data sparsity, interpretability, and transparent reasoning about demand drivers.

In brief

What This Work Does

This dissertation develops practical and interpretable ways to estimate who may use future air mobility services, where demand may emerge, and how data-driven forecasts can better support transportation planning.

Research impact

Why It Matters

Reliable demand estimates are essential for deciding where Advanced Air Mobility services and infrastructure should be deployed. By combining traveler behavior, spatial modeling, optimization, machine learning, and symbolic rules, this dissertation makes demand forecasts more useful and explainable for policy, infrastructure, and operational decisions.

Dissertation Contributions

1

Behaviorally Grounded Adoption

Develops a Generalized Cost of Trip framework incorporating monetary cost, travel time, reliability, and perceived risk across regional and urban case studies.

2

Data-Driven Spatial Estimation

Integrates learning with genetic optimization and gravity models to improve air mobility planning, nonlinear demand estimation, and flow representation.

3

Interpretable Neurosymbolic Learning

Embeds decision-tree-derived symbolic rules within neural networks to balance predictive accuracy, robustness, and transparent reasoning.

Key Contributions

  • Introduces a generalized-cost framework for modeling AAM adoption across multiple spatial scales.
  • Combines neural learning with classical optimization to accelerate airport-selection and flight-allocation planning.
  • Enhances gravity-based demand estimation with machine learning to capture nonlinear spatial and socioeconomic relationships.
  • Develops a neurosymbolic demand model that incorporates interpretable decision-tree rules into neural networks.

Evidence at a glance

Key Findings

Time + cost

Consistently emerge as the dominant factors shaping potential AAM adoption.

≈70% threshold

Adoption becomes substantially more likely when air travel accounts for more than approximately 70% of total generalized trip cost.

Spatial variation

Tennessee and New York City studies reveal that demand varies strongly across locations, trip contexts, and time periods.

Interpretable learning

Decision-tree rules embedded in neural networks support transparent reasoning and robustness when operational data are sparse.

Research connected to the dissertation

Selected Publications

Dissertation Committee

The committee brought together expertise in artificial intelligence, autonomous systems, transportation, software engineering, pervasive computing, and energy infrastructure.

Acknowledgment

Research Support

NASA Aeronautics Research Mission Directorate

University Leadership Initiative, Cooperative Agreement No. 80NSSC23M0059

U.S. National Science Foundation

Partial support through Grant Nos. 2309760 and 2317117

U.S. Department of Transportation

Center for Advanced Transportation Mobility, Grant No. 69A3551747125

Methods and Study Settings

Generalized Cost of Trip

Represents adoption decisions through monetary cost, time valuation, reliability, and perceived risk.

Spatial Case Studies

Evaluates Regional Air Mobility in Tennessee and high-resolution Urban Air Mobility demand across New York City taxi zones.

Learning-Accelerated Optimization

Uses neural networks to improve the efficiency of genetic-algorithm planning for airport selection and flight allocation.

Machine-Learning Gravity Model

Captures nonlinear spatial, socioeconomic, and transportation relationships missing from traditional gravity formulations.

Neurosymbolic Neural Networks

Integrates human-readable decision-tree rules with neural learning for interpretable and robust demand forecasts.

Publication Details

Type
Doctoral Dissertation
Year
2026
Degree
Doctor of Philosophy in Information Systems

Research Topics

Advanced Air Mobility Demand Modeling Neurosymbolic AI

Citation

@phdthesis{acharya2026enhancing,
  author={Acharya, Kamal},
  title={Enhancing Demand Modeling for Advanced Air Mobility Using Data-Driven Learning and Neurosymbolic AI},
  school={University of Maryland, Baltimore County},
  year={2026},
  type={Doctoral dissertation}
}