Perpustakaan Digital ITB

Flight Data Monitoring (FDM) is a system that plays a crucial role in aviation safety. This system enables continuous data collection and anomaly detection to identify risks and optimize operations. FDM is capable of collecting, analysing, and interpreting flight data recorded by aircraft systems to enhance flight safety and operational efficiency. FDM generates flight data containing various flight parameter information. This study uses data from a B747 aircraft, consisting of 241 records with the same origin and destination airports, as well as the same runway numbers for take off and landing. The research examines the application of machine learning, particularly the DBSCAN algorithm, in detecting anomalies within the FDM system. DBSCAN is an unsupervised learning method that can identify anomalous data without requiring predefined labels, making it suitable for complex flight data. DBSCAN detects anomalies by grouping data based on density, where data in low-density regions are considered anomalies. The research methodology consists of several key stages: data acquisition, preprocessing, machine learning model development, anomaly detection, validation and evaluation, and result interpretation. This method successfully detected 90% normal data, 5% anomalous data, and 5% outlier data caused by noise. The results indicate that DBSCAN effectively detects anomalies, making it a reliable solution for analysing high-dimensional flight data. This study concludes that applying DBSCAN can enhance aviation safety by enabling early anomaly detection and reducing operational risks. Further development is recommended to integrate DBSCAN with other machine learning methods to improve anomaly detection accuracy, optimize real-time data processing for faster anomaly detection, and evaluate the model on larger datasets to ensure its quality under various flight conditions. The utilization of machine learning for anomaly detection contributes to improving aviation safety and reliability.