Anomaly Detection in Time Series Flight Parameter Data Using Machine Learning Approach

Abstract

Abstract: Infrastructure monitoring is more crucial than ever, especially in the aviation industry. To attempt to overcome the difficulties brought on by the problems given by the exponential growth of connected devices and data volume, this study analyzes the application of machine learning approaches for anomaly identification in time series flight parameter data. The introduction of network telemetry, which automates data collecting, is presented as a remedy, however processing enormous data quantities in real-time still poses a challenge. With a focus on time-series data, the paper explores the role of machine learning in network telemetry anomaly detection. Statistical, proximity-based, deviation-based, and supervised classifiers are used to identify anomalies, or departures from predicted patterns, in flight parameter data. For a few examples of anomalous data, long short-term memory networks (LSTMs) are used. The objective is to provide an effective anomaly detection system that can process complex time series flight data and includes data purification, anomaly discovery, temporal reference, and value prediction. The methodology describes the univariate anomaly detection strategy, in which distinct models record particular patterns for each flight parameter. The findings offer new understanding of thresholds, repeated anomaly correction, and prediction errors. The results show that the method is accurate in separating instances of normal data from those containing anomalies, making it a useful tool for practical applications needing accurate anomaly identification.