Aircraft Flight Movement Anomaly Detection using Automatic Dependent Surveillance-Broadcast

Abstract

Automatic Dependent Surveillance-Broadcast (ADS-B) is an aircraft backup radar device that transmits aircraft sensor information via radio frequency. This data can be used to detect aircraft changes that occur significantly or abnormally (anomaly). Anomaly detection in this study aims to reduce and prevent flight accidents by analyzing abnormal data on aircraft flights using the Deep Learning (DL) model. Bidirectional LSTM (Bi-LSTM) and Bidirectional GRU (Bi-GRU) models are proposed as DL models which are implemented on ADS-B data using data mining methods. The data is generated from the ADS-B device that records the plane crash incident and is stored on the Flightradar24 community server. Data containing sensor changes from anomalous aircraft movements are studied for predictability on other flight data. The class breakdown is divided into two, anomaly and normal, based on information on the time span of anomaly occurrences in the accident investigation report of each aircraft using the sliding window technique in the data mining methodology. In evaluation, the confusion matrix measurement method is used to predict predictive analysis of the tested data. The results of the model evaluation performance show that the Bi-LSTM proposed in this study has the best accuracy of 99.44% and the f1-score of 99.51% is slightly superior to the Bi-GRU model. The model in this study can be applied in the ADS-B device to detect aircraft movement anomalies and as material for reviewing technicians in periodic maintenance and measuring the accuracy of the ADS-B device used as a backup radar.