Process Discovery using Rule-Integrated Trees Hidden Semi-Markov Models
| Autor: | Jia-Liang Guo, 郭家良 |
|---|---|
| Rok vydání: | 2017 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 105 To predict or to explain? With the dramatical growth of the volume of information generated from various information systems, data science has become popular and important in recent years while machine learning algorithms provide a very strong support and foundation for various data applications. Many data applications are based on black-box models. For example, a fraud detection system can predict which person will default but we cannot understand how the system consider it’s fraud. While white-box models are easy to understand but have relatively poor predictive performance. Hence, in this thesis, we propose a novel grafted tree algorithm to integrate trees of random forests. The model attempt to find a balance between a decision tree and a random forest. That is, the grafted tree have better interpretability and the performance than a single decision tree. With the decision tree is integrated from a random forest, it will be applied to Hidden semi-Markov models (HSMM) to build a Classification Tree Hidden Semi- Markov Model (CTHSMM) in order to discover underlying changes of a system. The experimental result shows that our proposed model RITHSMM is better than a simple decision tree based on Classification and Regression Trees and it can find more states/leaves so as to answer a kind of questions, “given a sequence of observable sequence, what are the most probable/relevant sequence of changes of a dynamic system?”. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|