Smart Recommendation for Anomaly Detection in IoT
Anomaly detection is the task of finding instances in a dataset that are different from the normal data. Today, anomaly detection is a core part of many IoT applications and finding abnormal instances is crucial in many applications. For example in network intrusion detection, identification of failures in mechanical systems and in smart sensors and abnormal usage of resources and diagnosis in the medical domain. In all of these applications, the amount of stored data has increased dramatically in the last decade, resulting in a strong demand for algorithms suitable for these crucial challenges. Existing research on anomaly detection has been fragmented across different application domains. Without a good understanding of how different techniques are related to each other and what the strengths and weaknesses of the techniques are, a large number of algorithms are created for the problem of anomaly detection in the field of IoT for specific use cases. This causes scalability issues in reusing the algorithm. In many solutions, the foundation of data are not taken into account, which leads to poor performance of the anomaly detection. The main goal of this thesis is to bridge this gap and to provide an efficient recommendation on anomaly detection algorithm, based on the good understanding of algorithms and characteristics of data. A broad spectrum of anomaly detection has been proposed mainly for semi-supervised and unsupervised anomaly detection. The assumptions, advantages, limitations, and variations are highlighted for algorithms and addressed for local, global and collective anomaly detection problems. The analysis of the characteristics of data is measured extensively, to resonance map the algorithms based on the characteristics and structure of the data. Hence this thesis proposes a formal solution, validated with the working prototype for recommending an anomaly detection algorithms. This solution allows the dynamic inspection of arbitrary IoT data in addition to an interactive environment to acquire domain knowledge when needed.
Darmstadt, TU, Master Thesis, 2019