Non-parametric Variable Selection for Interpretable Classification Models

Interpretable classification models require to refer only to a small subset of input variables; therefore, variable selection is essential. However, current variable selection methods for continuous inputs fail to achieve at least one of the requirements of being non-parametric, quantitative, conservative, and computationally efficient. This thesis develops an automatic variable selection framework that meets all these requirements for continuous variables in classification problems and evaluates it in a case study of modelling the outcomes of nano-material synthesis experiments.