Novelty based intrusion detection in unmanned aerial vehicles

This study aims to evaluate the performance of different machine learning methods for the detection of faults or anomalies due to novelty-based intrusions. Focusing on engine, aileron, rudder, and elevator faults, single-class-support vector machines (OC-SVM), single-class-random forest algorithm (OC-RF), local outlier factor algorithm (LOF) and autoencoder (Autoencoder) methods are used and the results are evaluated with accuracy, precision, sensitivity and F-score metrics. The OC-SVM method performed very well for engine, aileron, rudder and elevator faults. In this method, all performance metrics took the value of 1, providing high accuracy and efficiency in detecting faults. Similarly, the OC-RF method also showed a high performance in detecting engine, aileron, rudder and elevator failures. The fact that all metrics have a value of 1 indicates that this method is also a reliable option for fault detection. Although the LOF method showed a slightly lower performance compared to the other methods, it still gave successful results. The accuracy, sensitivity, precision, sensitivity and F-score metrics for engine, aileron, rudder and elevator faults have slightly lower values than the other methods. These values were obtained as 0.998. Autoencoder method obtained similar results with OC-SVM and OC-RF methods for all failure types. Accuracy, precision, sensitivity and F-score metrics were calculated as 1, thus proving the success and reliability of this method in fault detection. In conclusion, this study evaluated the effect of different machine learning methods for engine, aileron, rudder and elevator fault detection. The OC-SVM, OC-RF and Autoencoder methods performed well, while the LOF method had a slightly lower performance than the other methods. These results provide guidance in the evaluation and selection of different methods for fault detection. Future work can evaluate the impact of factors such as larger datasets and the addition of new features and contribute to the further development of fault detection systems. The thesis work can be a useful resource for researchers and industry experts working in the field of fault detection and can inspire further work. Keywords: Unmanned aerial vehicles, Novelty-based intrusion detection, Machine learning, One-class classifier.