Aviation Anomaly Detection Using Deep Convolutional Neural Network

Abstract

Due to aviation events caused by human mistakes, unwanted hardware failures, or criminal and terrorism intentions, a huge cost is incurred to governments annually. Due to the wider and more comprehensive tools, automatic detection methods can help us to reduce such problems considerably by analysis and evaluation of the events. These tools also play a key role in decreasing human errors and preventing terrorism and criminal cases. The present research evaluates the behavior of aircraft inside the airport to detect anomaly routes and situations. The proposed approach based on novel innovative Convolutional neural network, and autoencoder neural networks, can be implemented on airplanes with numerous arrival and departure lines, i.e. high-traffic airports such as John F. Kennedy airport. Abnormal behaviors identified using this scenario will play a significant role in reducing accidents within the airports. To assess the validity of the suggested method, this paper utilizes the dataset of aerial routes within the John F. Kennedy airport in the U.S. The method is compared with three conventional approaches in this field. Achieve to the 86 percent true positive rate and 89 percent area under the curve on test data, is an evidence for effectivity of method.