Abstract: |
This dissertation thesis has the following aims:
1. To map the already existing approaches to BN structure machine learning from data and the principles of their functioning with concentration on the approaches using genetic algorithms.
2. Based on this knowledge to propose the new, potentially more effective techniques for the BN structure learning from data.
3. To implement these proposed techniques (algorithms).
4. To design and perform the experiments on the benchmark (artificial) data.
5. To properly statistically analyse the results of those experiments.
6. To perform a case study on some real-world data.
7. To adumbrate the possible directions of the further research in the given area.
This dissertation thesis contains 11 chapters, which are organised as follows. Chapter 2 introduces the basic concepts needed: the uncertainty, probability, Bayesian network, the construction of BN, the missing values and the hidden variables, an inference in BN, main paradigms of expert systems and the current state of research in the given domain. In Chapter 3 data mining and the process of the automated BN construction from data are discussed. Chapter 4 provides an introduction to genetic algorithms. Chapter 5 deals with the issue of the use of GAs in machine learning of BNs from data. Chapter 6 presents and discusses new ideas and proposals relevant to the studied issue. Chapter 7 presents and discusses design of the experiments that have been performed in frame of this thesis. Chapter 8 presents and discusses more in detail results of the experiments performed on the artificial data. Chapter 9 applies the methods mentioned in the previous chapters on the real-world data coming from the educational area, more precisely the “Quality 2005” project, from SCIO. Chapter 10 concludes the discussion of the issue with the summary of the work that has been carried out and Chapter 11 offers several ideas about possible future developments in the area. |
