An Optimization of Classification Algorithms for Detecting Unauthorized Logins in Multi-Factor Authentication Environments

The objective of this research is to tackle the crimes of unauthorized logins in Multi-Factor Authentication systems with the help of supervised learning algorithms. Enhanced security is provided while still being prone to high level risks. This is the purpose of this work where Support Vector Machine, Random Forest, XGBoost and logistic regression are utilized for enhancing the performance by operating on false positive and false negative rates. On the other hand, most of the previous efforts have been of static nature, this study proposes another approach – rather than focusing on static assumptions it aims at detecting intrusions in the system in real time while the system is up and running. As a result, intruders can be detected in short time halving the time it takes to detect such intrusions with other security systems. The obtained results show that models which were created for real time detection have increased performance in this regard and while some gains in performance have been observed in many models the enhancements have been more pronounced in dynamic environments. This work contributes to the enhancement of the security of Multi-Factor Authentication systems and paves the way for further advancement in cyber security.