University of Twente Student Theses


IterSHAP: an XAI feature selection method for small high-dimensional datasets

Mourik, F.G. van (2023) IterSHAP: an XAI feature selection method for small high-dimensional datasets.

[img] PDF
Abstract:Small high-dimensional datasets pose challenges for achieving accurate predictive models, due to issues like overfitting and the curse of dimensionality. While complex models, like deep learning models, have been used to address these challenges, they often lack interpretability and transparency. Explainable Artificial Intelligence (XAI) is a popular field that aims to bridge this gap by developing techniques that provide insights into the decision-making process of machine learning models, therefore increasing the explainability and trustworthiness of models. However, when applying feature selection before model training, less-complex models can be used, such that interpretability and explainability are preserved, such that Explainable AI methods are not even needed. This research presents an iterative feature selection method named \textit{IterSHAP}, which utilizes a popular XAI technique named SHAP, to increase model performance on small high-dimensional datasets. The performance of IterSHAP was evaluated via both a simulation-based approach and an application-based approach. The results demonstrate the effectiveness of IterSHAP in selecting informative features and improving classification performance on small high-dimensional datasets. The limitations of IterSHAP are its convergence to a local optimum when working on large datasets and its lack of computational optimization.
Item Type:Essay (Master)
Slimstock, Deventer, The Netherlands
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science MSc (60300)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page