Designing an Estimation of Distribution Algorithm Based on Data Mining Methods

Akbari Azirani, Elham | 2013

493 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 45320 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Beigy, Hamid
  7. Abstract:
  8. Estimation of distribution algorithms (EDA) are optimization methods that search the solution space by building and sampling probabilistic models. The linkage tree genetic algorithm (LTGA), which can be considered an estimation of distribution algorithm, uses hierarchical clustering to build a hierarchical linkage model called the linkage tree, and gene-pool optimal mixing algorithm to generate new solutions. While the LTGA performs very well on problems with separable sub-problems, its performance deteriorates on ones with overlapping sub-problems. This thesis presents a comparison of the effect of different pre-constructed models in the LTGA's performance. Several important factors that increase the quality of the model are derived from this comparison. Two LTGA-based algorithms are presented with regard to these factors: First, the Overlapping-LTGA which alters the clustering method, enabling it to build a broader class of models that unlike the linkage tree, allow some degree of overlap between the linkage groups. Second, the Pruned-LTGA, which filters out the linkage tree based on linkage group size. Experiments show that both the algorithms improve the LTGA in some ways. The thesis also offers an introduction to EDAs and a review of discrete EDAs
  9. Keywords:
  10. Distribution Estimation Algorithm ; Mutual Information ; Hierarchical Clustering ; Optimal Mixing Evolutionary Algorithm ; Overlapping Sub-Problems ; Subjects Family ; Linkage Tree Genetic Algorithm (LTGA)

 Digital Object List