Shapley | DICE | Explainable AI | Classifier | EGBoost
Predicting Student Employability and Personalizing Advice Through Machine Learning Using Mock Interview Results
This paper was presented at the 2024 Asian Institute of Management Machine Learning 2 public presentation event, together with my classmates and colleagues, Ian CoKehyeng, Rex Laylo, Gregory Uy, and John Cris Orenday.
Using Random Undersampling and an XGBoost Classifier model, we showed that it possible to predict the employability assessment given to Technological Instutute of the Philippines students by their OJTs using only mock-interview results at an 85% accuracy.
Furthermore, we demonstrated using DICE Counterfactual Explanations how a students’ employability score may be improved through the smallest change in soft skills as assessed by the mock interviews. For example, if this student increases their “general appearance” score, they’re much more likely to be assessed as emplyable.
Finally, we found that as few as 500 mock interview results, a feasible number for a university to collect, already achieves an 80% accuracy.
The full notebook can be accessed here, and the paper can be accessed here.