Salwani Abdullah

A Hybrid Harmony Search Algorithm for Solving Dynamic Optimisation Problems

Many optimisation problems are dynamic in the sense that changes occur during the optimisation process, and therefore are more challenging than the stationary problems. To solve dynamic optimisation problems, the proposed approaches should not only attempt to seek the global optima but be able to also keep track of changes in the track record of landscape solutions. In this research work, one of the most recent new population-based meta-heuristic optimisation technique called a harmony search algorithm for dynamic optimization problems is investigated. This technique mimics the musical process when a musician attempts to find a state of harmony. In order to cope with a dynamic behaviour, the proposed harmony search algorithm was hybridised with a (i) random immigrant, (ii) memory mechanism and (iii) memory based immigrant scheme. The performance of the proposed harmony search is verified by using the well-known dynamic test problem called the Moving Peak Benchmark (MPB) with a variety of peaks. The empirical results demonstrate that the proposed algorithm is able to obtain competitive results, but not the best for most of the cases, when compared to the best known results in the scientific literature published so far.

Fuzzy Modified Great Deluge Algorithm for Attribute Reduction

This paper proposes a local search meta-heuristic free of parameter tuning to solve the attribute reduction problem. Attribute reduction can be defined as the process of finding minimal subset of attributes from an original set with minimum loss of information. Rough set theory has been used for attribute reduction with much success. However, the reduction method inside rough set theory is applicable only to small datasets, since finding all possible reducts is a time consuming process. This motivates many researchers to find alternative approaches to solve the attribute reduction problem. The proposed method, Fuzzy Modified Great Deluge algorithm (Fuzzy-mGD), has one generic parameter which is controlled throughout the search process by using a fuzzy logic controller. Computational experiments confirmed that the Fuzzy-mGD algorithm produces good results, with greater efficiency for attribute reduction, when compared with other meta-heuristic approaches from the literature.

Nature-Inspired Chemical Reaction Optimisation Algorithm for Handling Nurse Rostering Problem

The optimisation of the nurse rostering problem is chosen in this work seeking to improve the organization of hospital duties and to elevate health care by enhancing the quality of the decision-making process. Nurse rostering is a difficult and complex problem with a large number of demands and requirements that conflict with hospital workload constraints in terms of employee work regulations and personal preferences. We propose a variable population-based metaheuristic algorithm, the chemical reaction optimisation (CRO), to solve the NRP at the First International Nurse Rostering Competition (2010). The CRO algorithm features an adaptive search procedure that systematically controls the selection between an intensive search strategy and diversification search based on specific criteria to reach the best solution. Computational results were measured with three complexity levels as a total of 30 variant instances based on real-world constraints.

Comparison between compactness and connectedness criteria in data clustering

Data clustering is the first step in data mining. It aims at finding homogeneous groups of objects based on the degree of similarity and dissimilarity of their attributes. Most of the existing clustering methods are based on a single criterion to measure the goodness of clusters. In most cases, these methods are not suitable for different types of datasets with different characteristics. In this study, biogeography-based optimisation BBO and great deluge GD algorithms are combined to address the data clustering as single objective optimisation problem; two versions of the proposed approach that employed two different clustering criteria as the objective function have been investigated using fourteen 2D synthetic benchmark datasets. The quality of the obtained clusters of both versions of the proposed approach is insufficient with respect to the external evaluation function i.e. F-measure. Thus, the data-clustering problem preferred to be tackled as multi-objective clustering algorithms.

Intelligent Double Treatment Iterative Algorithm for Attribute Reduction Problems

Attribute reduction is a combinatorial optimization problem in data mining that aims to find minimal reducts from large set of attributes. The problem is exacerbated if the number of instances is large. Therefore, this paper concentrates on a double treatment iterative improvement algorithm with intelligent selection on composite neighbourhood structure to solve the attribute reduction problems and to obtain near optimal reducts. The algorithm works iteratively with only accepting an improved solution. The proposed approach has been tested on a set of 13 benchmark datasets taken from the University of California, Irvine (UCI) machine learning repository in line with the state-of-the-art methods. The 13 datasets have been chosen due to the differences in size and complexity in order to test the stability of the proposed algorithm. The experimental results show that the proposed approach is able to produce competitive results for the tested datasets.

Hybrid Radial Basis Function with Particle Swarm Optimisation Algorithm for Time Series Prediction Problems

Time Series Prediction (TSP) is to estimate some future value based on current and past data samples. Researches indicated that most of models applied on TSP suffer from a number of shortcomings such as easily trapped into a local optimum, premature convergence, and high computation complexity. In order to tackle these shortcomings, this research proposes a method which is Radial Base Function hybrid with Particle Swarm Optimization algorithm (RBF-PSO). The method is applied on two well-known benchmarks dataset Mackey-Glass Time Series (MGTS) and Competition on Artificial Time Series (CATS) and one real world dataset called the Rainfall dataset. The results revealed that the RBF-PSO yields competitive results in comparison with other methods tested on the same datasets, if not the best for MGTS case. The results also demonstrate that the proposed method is able to produce good prediction accuracy when tested on real world rainfall dataset as well.

Variable Neighbourhood Iterated Improvement Search Algorithm for Attribute Reduction Problems

Attribute reduction is one of the main contributions in Rough Set Theory RST that tries to find all possible reducts by eliminating redundant attributes while maintaining the information of the problem in hand. In this paper, we propose a meta-heuristic approach called a Variable Neighbourhood Iterated Improvement Search VNS-IIS algorithm for attribute reduction. It is a combination of the variable neighbourhood search with the iterated search algorithm where two local search algorithms i.e. a random iterated local search and a sequential iterated local search algorithm are employed in a parallel strategy. In VNS-IIS, an improved solution will always be accepted. The proposed method has been tested on the 13 well-known datasets that are available in the UCI machine learning repository. Experimental results show that the VNS-IIS is able to obtain competitive results when compared with other approaches mentioned in the literature in terms of minimal reducts.

Electromagnetic algorithm for tuning the structure and parameters of neural networks

Electromagnetic algorithm is a population based meta-heuristic which imitates the attraction and repulsion of sample points. In this paper, we propose an electromagnetic algorithm to simultaneously tune the structure and parameter of the feed forward neural network. Each solution in the electromagnetic algorithm contains both the design structure and the parameters values of the neural network. This solution later will be used by the neural network to represents its configuration. The classification accuracy returned by the neural network represents the quality of the solution. The performance of the proposed method is verified by using the well-known classification benchmarks and compared against the latest methodologies in the literature. Empirical results demonstrate that the proposed algorithm is able to obtain competitive results, when compared to the best-known results in the literature.