Lydia Wahid Amin Rizkallah

Network intrusion detection is the problem of detecting suspicious requests through networks. In recent years, many researchers focus on addressing this problem in the context of machine learning. Although machine learning algorithms are powerful, most of them lack the power of interpretability. Expert systems, on the other hand, are knowledge-based systems designed to simulate the problem-solving behavior of human experts. Expert systems possess the advantage of interpretability through an explanation mechanism that justifies its own line of reasoning, however, they need the availability of a domain expert. This paper proposes the use of rule learning approaches to gain the best of both fields, being interpretable as expert system and learnable through collected datasets without the need for explicit expertise. A separate and conquer rule learning approach is proposed for network intrusion detection. Our results show that the separate and conquer approach achieves a 0.99 weighted average F1-score on the test set which makes it very comparative to both decision trees and classical machine learning approaches. We also show that rules produced using separate and conquer are much simpler than decision trees and more interpretable.

For any college or university student, selecting the right courses in which to enrol is a critical choice that has the potential to positively or negatively affects the student's academic performance. For this reason, universities offer academic advisors, which study the case of each student and recommend the right courses based on the student's status. This process is expensive and time-consuming, therefore, researchers have proposed solutions to automate it. In this paper, we focus on solutions based on expert systems. To this end, we summarize and analyze six selected works that propose different expert systems to solve the problem of academic advising. The analysis aims to provide an overview about the used approaches as well as highlight areas of improvements. The goal is to help any researcher interested in the problem to take the first step towards learning about existing approaches, their advantages, and their drawbacks. Overall, existing approaches show great potential of expert systems in the problem of academic advising.