Traditional clustering algorithms have different criteria and biases, and there is no single algorithm that can be the best solution for a wide range of data sets. This problem often presents a significant obstacle to analysts in revealing meaningful information buried among the huge amount of data. Ensemble Clustering has been proposed as a way to avoid the biases and improve the accuracy of clustering. The difficulty in developing Ensemble Clustering methods is to combine external information (provided by input clusterings) with internal information (i.e. characteristics of given data) effectively to improve the accuracy of clustering. The work presented in this thesis focuses on enhancing the clustering accuracy of Ensemble Clustering by employing heuristic optimisation techniques to achieve a robust combination of relevant information during the consensus clustering stage. Two novel heuristic optimisation-based Ensemble Clustering methods, Multi-Optimisation Consensus Clustering (MOCC) and K-Ants Consensus Clustering (KACC), are developed and introduced in this thesis. These methods utilise two heuristic optimisation algorithms (Simulated Annealing and Ant Colony Optimisation) for their Ensemble Clustering frameworks, and have been proved to outperform other methods in the area. The extensive experimental results, together with a detailed analysis, will be presented in this thesis.

In previous studies it was determined that the fiber diameter distribution in a peripheral nerve could be estimated by a simulation technique known as group delay. These results could be further improved using a combinatorial optimization algorithm called simulated annealing. This paper explores the structure and behavior of simulated annealing for the application of optimizing the group delay estimated fiber diameter distribution. Specifically, a set of parameters known as the cooling schedule is investigated to determine its effectiveness in the optimization process. Simulated annealing is a technique for finding the global minimum (or maximum) of a cost function which may have many local minima. The set of parameters which comprise the cooling schedule dictate the rate at which simulated annealing reaches its final solution. Converging too quickly can result in sub-optimal solutions while taking too long to determine a solution can result in an unnecessarily large computational effort that would be impractical in a real-world setting. The goal of this study is to minimize the computational effort of simulated annealing without sacrificing its effectiveness at minimizing the cost function. The cost function for this application is an error value computed as the difference in the maximum compound evoked potentials between an empirically-determined template distribution of fiber diameters and an optimized set of fiber diameters. The resulting information will be useful when developing the group delay estimation and subsequent simulated annealing optimization in an experimental laboratory setting.