In the scope of this project, we develop various cooperative-game theory-based explainable AI methods to evaluate the feature importance values for a regression model with a multicollinearity issue. This issue prevents describing the relationship between the features and the target accurately. We also look at the legal aspects, when the model does not generate accurate predictions.
In this project, we aimed to compare the performance of human participants and software agents in forming the ideal coalition. To do this, we conducted human subject experiments to collect data and developed a similar setting for the software agents. Our focus was on the “glove game, a commonly used cooperative game in human experiments. Our results showed that the software agents performed similarly to human players in terms of finding the optimal coalition. Specifically, we evaluated two trials: one with only software agents and another with a combination of human and sonftware agents. Overall, our study suggests that software agents can be effective in forming coalitions in cooperative games.
This project analyzed the impact of cyberloafing on an organization’s cyber risk. Employee behavior, specifically cyberloafing, affects productivity and creates an opportunity for malware to be introduced into the corporate system. Factors such as productivity, workload, and corporate sanctions have varying effects on cyberloafing and therefore on the cyber risk. While sanctions can help reduce cyberloafing, factors related to workload have a greater effect on employee tendencies toward cyberloafing and subsequently on the organization’s cyber risk.
This project analyzed the application of game theory to address systems engineering problems. Different engineering journals were reviewed to understand the game-theoretic models and the engineering problems discussed. This work is a great introduction to the basics of game theory and includes some examples of non-cooperative and cooperative game theoretic models.