Computer Science Department Thesis Defense - Liam Dingle

Please join the Computer Science Department for the upcoming thesis defense:

Presenter: Liam Dingle

Thesis title: A Comparison of Explainable Clinical Decision Making Using White Box and Black Box Models

Abstract: Explainability is a crucial element of machine learning based decision making in high stake scenarios. The area of ICU outcome prediction is one such area. There currently exists a performance tradeoff between low-complexity machine learning models capable of making predictions that are inherently interpretable (white box) to a human, and cutting-edge high complexity (black box) models that fit better to non-linear decision boundariesare not readily interpretable . This document aims to assess the reliability of the predictions made by black-box models by comparing the decisions made by white box models and their black box counterparts by contrasting explainable model feature coefficients/importances to feature importance values generated by a post-hoc SHAP values. The comparison between explainable models and non-explainable models show that both series of models prioritize clinically relevant variables when making outcome predictions. Then in another study, we test the reliability of the generated explanations made by both SHAP and explainable model importances by assessing the performance impact of training models on different subsets of features generated. This is performed by ranking features based on each type of feature importance. To introduce further rigor, two other binary classification tasks are performed on two different clinical datasets . This analysis shows that there is a tangible performance impact between decisions that can be correctly made on important compared to unimportant data. Overall, we can conclude that the implementation of black box models in high-stakes decision making can offer tangible benefits in performance while also providing reliable, transparent predictions if proper explainability mechanisms are in place.

Committee Members:
Dr. Yimin Yang (supervisor, committee chair), Dr. Ruizhong Wei (co-supervisor), Dr. Amin Safaei, Dr. Thangarajah Akilan (Software Engineering)

Please contact grad.compsci@lakeheadu.ca for the Zoom link. Everyone is welcome.