Probabilistic Smoothing for Tree and Rule Ensemble Methods
Year: 2026
Type: dissertation
Abstract: This thesis develops a framework for making predictions from tree-based machine learning models smoother and more reliable. Standard methods of this type produce uneven prediction surfaces and poorly estimated uncertainties, limiting their use in risk-sensitive decisions. This thesis applies kernel smoothing to two model families — random forests and rule ensembles — to address both problems simultaneously. The approach produces well-calibrated uncertainty estimates across varying data sizes, while rule ensembles retain their interpretability. Testing on real-world datasets shows consistent and stable performance. This work demonstrates that smoothing is a practical and principled way to improve the reliability of tree-based machine learning models.
Source: Monash University
DOI: 10.26180/33697057
Author: Ziyi Liu
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7212431386
