Abstract
This thesis aimed to answer the question: can domain knowledge regarding nondecreasing (monotone) relationships be incorporated into high performance classification algorithms without compromising their performance? Two high accuracy classifiers (Support Vector Machines and Random Forest) and a foundational approach (instance based partial orders) were extended into novel versions that demonstrated improvements in one or more of the following: global monotonicity (guaranteeing compliance in all circumstances), partial monotonicity (allowing non-monotone features), prediction accuracy, and scalability. It is hoped this thesis encourages and enables practitioners to include knowledge of monotonicity in their models, for the sake of accuracy, simplicity, and interpretability.
Original language | English |
---|---|
Qualification | Doctor of Philosophy |
Awarding Institution |
|
Thesis sponsors | |
Award date | 13 Jun 2019 |
DOIs | |
Publication status | Unpublished - 2018 |