Review of ADS-B Data Usage with the focus on Data Cleaning

Abstract

Automatic Dependent Surveillance–Broadcast (ADS-B) data have become a vital resource for research on trajectory prediction, conflict detection, and air traffic management. However, due to limitations in data acquisition and transmission, ADS-B datasets often contain missing points, irregular sampling, and anomalies. To ensure usability, researchers typically apply data cleaning and preprocessing, which improve data quality but may alter original characteristics and cause deviations between algorithm outputs and real operational patterns. Existing studies largely focus on individual cleaning methods, lacking systematic and quantitative assessments of their impact on downstream applications. To address this gap, this study systematically investigates the relationship between data cleaning and algorithmic performance in ADS-B analytics. It reviews major ADS-B applications and prevalent cleaning techniques, summarizes typical preprocessing pipelines, and provides guidance for building more robust evaluation frameworks. An AutoEncoder (AE)-based experiment is conducted using three architecturally distinct autoencoders (Fully Connected AE (FC-AE), Long Short-Term Memory AE (LSTM-AE), Gated Recurrent Unit AE (GRU-AE)) across four geographically diverse airport datasets, with trajectories contaminated by Gaussian, drift, spike, and missing data noise to assess the influence of cleaning strategies on trajectory reconstruction. Results indicate that spike noise has the least impact on reconstruction performance across all model and dataset combinations, while the relative sensitivity to Gaussian, drift, and missing data noise varies primarily with dataset characteristics rather than model architecture. These findings suggest that data cleaning priorities should be informed by the specific noise profile of the operational environment rather than applied uniformly.