Emin Erkan Korkmaz

An experimental study on hyper-heuristics and exam timetabling

Hyper-heuristics are proposed as a higher level of abstraction as compared to the metaheuristics. Hyper-heuristic methods deploy a set of simple heuristics and use only non-problem-specific data, such as fitness change or heuristic execution time. A typical iteration of a hyper-heuristic algorithm consists of two phases: the heuristic selection method and move acceptance. In this paper, heuristic selection mechanisms and move acceptance criteria in hyper-heuristics are analyzed in depth. Seven heuristic selection methods and five acceptance criteria are implemented. The performance of each selection and acceptance mechanism pair is evaluated on 14 well-known benchmark functions and 21 exam timetabling problem instances.

Linear linkage encoding in grouping problems: applications on graph coloring and timetabling

Linear Linkage Encoding (LLE) is a recently proposed representation scheme for evolutionary algorithms. This representation has been used only in data clustering. However, it is also suitable for grouping problems. In this paper, we investigate LLE on two grouping problems; graph coloring and exam timetabling. Two crossover operators suitable for LLE are proposed and compared to the existing ones. Initial results show that LLE is a viable candidate for grouping problems whenever appropriate genetic operators are used.

Multi-objective Genetic Algorithms for grouping problems

Linear Linkage Encoding (LLE) is a convenient representational scheme for Genetic Algorithms (GAs). LLE can be used when a GA is applied to a grouping problem and this representation does not suffer from the redundancy problem that exists in classical encoding schemes. LLE has been mainly used in data clustering. One-point crossover has been utilized in these applications. In fact, the standard recombination operators are not suitable to be used with LLE. These operators can easily disturb the building blocks and cannot fully exploit the power of the representation. In this study, a new crossover operator is introduced for LLE. The operator which is named as group-crossover is tested on the data clustering problem and a very significant performance increase is obtained compared to classical one-point and uniform crossover operations. Graph coloring is the second domain where the proposed framework is tested. This is a challenging combinatorial optimization problem for search methods and no significant success has been obtained on the problem with pure GA. The experimental results denote that GAs powered with LLE can provide satisfactory outcomes in this domain, too.

Novel Clustering Approach that Employs Genetic Algorithm with New Representation Scheme and Multiple Objectives

In this paper, we propose a new encoding scheme for GA and employ multiple objectives in handling the clustering problem. The proposed encoding scheme uses links so that objects to be clustered form a linear pseudo-graph. As multiple objectives are concerned, we used two objectives: 1) to minimize the Total Within Cluster Variation (TWCV); and 2) minimizing the number of clusters in a partition. Our approach obtains the optimal partitions for all the possible numbers of clusters in the Pareto Optimal set returned by a single GA run. The performance of the proposed approach has been tested using two well-known data sets: Iris and Ruspini. The obtained results demonstrate improvement over classical approaches.

A Grouping Genetic Algorithm Using Linear Linkage Encoding for Bin Packing

Linear Linkage Encoding (LLE) is a representation method proposed for grouping problems. It has already been used in solving data clustering, graph coloring and timetabling problems based on multi-objective genetic algorithms. In this study, this novel encoding scheme is investigated on bin packing again using a genetic algorithm. Bin packing benchmark problem instances are used to compare the performance of traditional recombination operators and custom made LLE crossover operators which are hybridized with parametrized placement heuristics. The results denote that LLE is a viable candidate for bin packing problem whenever appropriate genetic operators are chosen.

A controlled genetic programming approach for the deceptive domain

Traditional genetic programming (GP) randomly combines subtrees by applying crossover. There is a growing interest in methods that can control such recombination operations in order to achieve faster convergence. In this paper, a new approach is presented for guiding the recombination process for genetic programming. The method is based on extracting the global information of the promising solutions that appear during the genetic search. The aim is to use this information to control the crossover operation afterwards. A separate control module is used to process the collected information. This module guides the search process by sending feedback to the genetic engine about the consequences of possible recombination alternatives.

Effective data mining by integrating genetic algorithm into the data preprocessing phase

Dividing a data set into a training set and a test set is a fundamental component in the preprocessing phase of data mining (DM). Effectively, the choice of the training set is an important factor in deriving good classification rules. Traditional approach for association rules mining divides the dataset into training set and test set based on statistical methods. In this paper, we highlight the weaknesses of the existing approach and hence propose a new methodology that employs genetic algorithm (GA) in the process. In our approach, the original dataset is divided into sample and validation sets. Then, GA is used to find an appropriate split of the sample set into training and test sets. We demonstrate through experiments that using the obtained training set as the input to an association rules mining algorithm generates high accuracy classification rules. The rules are tested on the validation set for accuracy. The results are very satisfactory; they demonstrate the applicability and effectiveness of our approach.

Design and Usage of a New Benchmark Problem for Genetic Programming

Not so many benchmark problems have been proposed in the area of Genetic Programming (GP). In this study, a new artificial benchmark problem is designed for GP. The different parameters that can be used to tune the difficulty of the problem are analyzed. Also, the initial experimental results obtained on different instances of the problem are presented.

Hybrid local search algorithms on Graph Coloring Problem

Hybridization of local search algorithms yield promising algorithms for combinatorial optimization problems such as Graph Coloring Problem (GCP). This paper presents a new meta-heuristic Simulated Annealing with Backtracking (SABT) and shows the effect of hill climber and tabu search on SABT for solving GCP. The algorithm proposed merges the power of simulated annealing approach and backtracking mechanism. Some hill climbers are integrated for fine tuning and also tabu search is integrated for avoiding from redundant search. Several tests are run on a collection of benchmarks from DIMACS challenge suite and promising results are obtained. A comparison of SABT framework with some other state-of-the-art algorithms is presented along with an analysis of the performance of the algorithm.

Representation issue in graph coloring

Linear Linkage Encoding (LLE) is a powerful encoding scheme utilized when genetic algorithms (GAs) are applied to grouping problems. It discards the redundancy of other traditional encoding schemes. However, some genetic operators are quite costly in terms of computational time when LLE is utilized. In this study, two supplementary encoding schemes Linear Linkage Encoding with Ending Node Links (LLE-e) and Linear Linkage Encoding with Backward Links (LLE-b) are designed and used together with LLE. When a genetic operator is costly in LLE, the operation is carried out on one of the supplementary encodings and the result is reflected back to LLE. The algorithm is implemented in a Multi Objective Genetic Algorithm (MOGA) framework and various tests are carried out on Graph coloring Problem (GCP) instances obtained from DIMACS Challenge Suite. A performance improvement has been obtained when both supplementary encoding schemes are included in the algorithm.