Kamal Acharya

Postdoctoral Research Associate at Baylor University

Back to Publications

Conference Paper

Enhancing Forecasting for Advanced Air Mobility

By Mehul Lad Kamal Acharya Liang Sun Houbing Herbert Song

An IEEE BigData 2024 paper on hourly flight-demand forecasting for Advanced Air Mobility using temporal features and machine learning models.

2024 IEEE BigData 2024 DOI: 10.1109/BigData62323.2024.10825414

Advanced Air Mobility Forecasting Big Data

DOI Cite

Abstract

Accurately predicting flight demand is essential for optimizing air travel operations and resource allocation. In our research, we explore the relationship between temporal patterns and flight demand, leveraging hourly data rather than traditional meteorological factors. Through analysis, we discovered significant correlations between hour of the day and flight demand, prompting the creation of features such as peak hours and time segments (morning, afternoon, evening). By utilizing these temporal features, we develop predictive models employing various machine learning algorithms, including LSTM, linear regression, and gradient boosting models. We aim to identify the most effective approach for accurately forecasting flight demand, with implications extending to the optimization of Advanced Air Mobility (AAM) solutions, where understanding temporal patterns is crucial for efficient resource allocation and urban air transportation network design.

In brief

What This Work Does

The paper predicts airport flight demand from hourly patterns so future AAM systems can schedule vehicles and allocate resources more effectively.

Research impact

Why It Matters

Reliable forecasting is necessary before AAM services can be planned, evaluated, and deployed at scale.

Paper at a glance

Research Scope

4Research stages
4Methods or application areas
5Future research directions

Hourly Flight-Demand Forecasting Workflow

1

Collect Data

Combines NOAA hourly weather observations with FAA records of flight departures and arrivals.

2

Engineer Temporal Features

Creates hour-of-day, peak-hour, and time-segment features to represent demand patterns.

3

Train Forecasting Models

Fits LSTM, linear regression, gradient boosting, and transformer models to historical flight-demand data.

4

Evaluate Forecasts

Compares models using MSE, MAE, MAPE, and R-squared across selected Tennessee airports.

Key Contributions

  • Models flight demand using hourly temporal patterns rather than relying only on meteorological or seasonal factors.
  • Creates features such as peak hours and time-of-day segments to capture operational demand variation.
  • Compares LSTM, linear regression, gradient boosting, and transformer-based approaches.
  • Evaluates forecasting performance across multiple Tennessee airports.
  • Connects temporal flight-demand prediction to AAM resource allocation and network planning.

Forecasting Methods

LSTM

Captures long-term dependencies in hourly flight-demand sequences.

Transformer

Models complex temporal sequence patterns and shows strong reported MAPE performance.

Gradient Boosting

Builds predictive power through ensembles of weak learners over engineered demand features.

Linear Regression

Provides an interpretable baseline for comparing more expressive time-series models.

Research Gaps

  1. Hourly demand variation
  2. Temporal feature design
  3. Airport-level transferability
  4. AAM resource allocation
  5. Network scheduling

Publication Details

Type
Conference Paper
Venue
IEEE BigData 2024
Year
2024
Pages
2908-2913

Research Topics

Advanced Air Mobility Forecasting Big Data Flight Demand Time-Series Forecasting

Citation

@inproceedings{lad2024enhancing,
  author={Lad, Meet and Acharya, Kamal and Sun, Liang and Song, Houbing},
  title={Enhancing Forecasting for Advanced Air Mobility},
  booktitle={IEEE BigData 2024},
  year={2024},
  pages={2908--2913},
  doi={10.1109/BigData62323.2024.10825414}
}