University of North Texas
Enhancing Interpretability of Convolutional Neural Networks: A Decision Tree Approach Using Grad-CAM and Wavelet Transforms for Multi-Class Classification
Abstract
dc:descriptionThis research extends previous work on interpreting Convolutional Neural Networks (CNNs) to apply improved techniques to understand the model’s ability to make decisions in multi-class classification tasks. Classification by CNNs was explained in the original study using feature map extraction and decision trees. In this work, a technique known as Grad-CAM (Gradient-weighted Class Activation Mapping) is used to obtain heat maps that illustrate regions in the image responsible for the CNN class distinction. To increase the readability of these heatmaps, wavelet transforms are used to denoise the cropped images prior to the use of Grad-CAM. It is then followed for construction of a decision tree by extracting features; area, perimeter, and aspect ratio from heatmaps. It lies in this approach of offering insightful understanding of the CNN’s decision and its rationale for categorizing image data into six different classes. Here I attempt to provide some additions to the current body of model interpretability research by focusing on the behavior of CNNs and by laying groundwork for more practical and effective methods of explaining intricate machine learning.
Degree
thesis:*- Grantor dc:publisher
- University of North Texas
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Umesha, Jeevan
- Contributors dc:contributor
-
- Bryce, Renee
- Feng, Yunhe
- Pears, Russel
Subjects
dc:subject × 5Rights
dc:rights- Statement dc:rights
-
- Public
- Umesha, Jeevan
- Copyright
- Copyright is held by the author, unless otherwise noted. All rights Reserved.
- Language dc:language
- English
Identifiers
dc:identifier.*- Identifier
-
local-cont-no: submission_4147
https://digital.library.unt.edu/ark:/67531/metadc2443111/
ark: ark:/67531/metadc2443111