Detection of Suspicious Behavior, through Machine Learning, on the applicative layer and data records

The rapid proliferation of bots in online environments poses a significant threat to the integrity and functionality of various platforms, ranging from social media networks to e-commerce websites. Currently, one of the most threatening bot attacks for airlines is Seat-Spinning - a fraudulent technique that exploits bots to initiate DoI attacks, resulting in significant losses for airline companies. The scope of this thesis is to propose an alerting system capable of detecting such attacks by leveraging flight inventory data. The problem will be modeled as an anomaly detection in time series in an online framework. The proposed pipeline utilizes reconstruction models, dynamic thresholding, and window scoring to perform the detection task on streaming data. Different solutions have been evaluated for the problem by employing NAB scoring and considering time resources. This process aids us in identifying the fastest and most efficient detector for the task. Finally, the solution has been validated in a live-streaming environment, showcasing the effectiveness of this detection system in real-time inventory monitoring. Despite the potential for numerous enhancements and system integrations, this thesis demonstrates that an ML-based solution for detecting bot traffic engaged in DoI attacks is both feasible and effective. This approach offers a substantial advantage in countering malicious threats in online environments and poses a solid basis for future research.