Machine Learning Systems Rationale Analysis

Wahl, Timo (2021) Machine Learning Systems Rationale Analysis. Bachelor's Thesis, Artificial Intelligence.

Preview	Text bAI_2021_WahlT.pdf Download (592kB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (119kB)

Abstract

Bench-Capon (1993) showed that high accuracies on an open texture classification task does not necessarily mean that the reasoning of the model making the prediction is sound. In that paper it was shown that an in-depth analysis of the decision making process of an MLP can indicate the discrepancies in the models understanding, or rationale, of the domain that was trained on. This paper first replicated the original paper by Bench-Capon and then extended that paper with a more in-depth analysis of the domain, new machine learning systems and improved hyperparameter optimization. The MLP, Random Forest classifier and XGBoost classifier were all compared in their performance and rationales on the same domain. The conclusions from the paper by Bench-Capon were reaffirmed, with the primary conclusion from the extension being that the Random Forest and XGBoost classifiers are not capable of learning completely sound rationales for the welfare domain, despite being more optimized and being trained on larger datasets.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Steging, C.C. and Verheij, H.B.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	01 Jul 2021 10:28
Last Modified:	01 Jul 2021 10:28
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/24844

Actions (login required)

View Item