Type
Text
Type
Dissertation
Advisor
Wu, Song | Zhu, Wei | Gao, Yi | Xiao, Keli.
Date
2016-12-01
Keywords
Applied mathematics | classification algorithm, classifier combination, computer-aided diagnosis, location index, machine learning
Department
Department of Applied Mathematics and Statistics
Language
en_US
Source
This work is sponsored by the Stony Brook University Graduate School in compliance with the requirements for completion of degree.
Identifier
http://hdl.handle.net/11401/77155
Publisher
The Graduate School, Stony Brook University: Stony Brook, NY.
Format
application/pdf
Abstract
Machine learning addresses the question of how computer make decisions and predictions automatically through existing experiences and data, which has become an increasingly important topic with the advent of modern data science and automated big data analysis. Several algorithms are widely used in machine learning. However, each classifier, inevitably, has certain inadequacy for which we hope to compensate. To address these issues, this study first introduces the necessary theoretical background and principles for machine learning and those typical classifiers. Based on these classifiers, this paper attempts to (1) use bagging/boosting to improve the simple classifier, and, (2) find some combination strategies to make use of the advantage of each classifier. The second part of this paper is to verify the robustness of these innovative ideas via multiple datasets. First, several common datasets are analyzed with the results compared between our new algorithm and those typical classifiers. Overall, we can obtain some gains in terms of the AUC value in virtually every dataset with the new algorithm and significant gains in most dataset. Secondly, we apply these algorithms to a real-life image data classification problem. The pipeline of this project includes 3D texture feature amplification, feature extraction via KL-transform, feature selection and classification. Finally, we gladly report that significant improvements have been achieved through both the new feature selection method and the new classification algorithm. | 78 pages
Recommended Citation
Hu, Yifan, "Exploration of statistical learning strategies and their applications on medical image data for computer-aided diagnosis" (2016). Stony Brook Theses and Dissertations Collection, 2006-2020 (closed to submissions). 2991.
https://commons.library.stonybrook.edu/stony-brook-theses-and-dissertations-collection/2991