Knowledge discovery is one of the most important human activities, which helps people recognise and understand some of the intricacies associated with the ancient and modern worlds. With the rapid development in the human capabilities to both generate and collect data, the discovery of knowledge from data has become a practical and popular research topic. In this thesis, knowledge discovery from data is conducted from the following two overarching viewpoints: first, developing prediction models using the data that represent input-output relationships; second, based on these developed prediction models, finding the optimal designs (solutions) from a set of predefined objectives. The theoretical aspects behind the previous two research facets are described and the associated experimental studies are carried out.
A particular focus of this thesis is on a cooperative fuzzy modelling framework, which integrates transparent (interpretable) fuzzy systems with robust evolutionary computing based algorithms involving several techniques, such as data clustering, data mining, and multi-objective optimisation. Evolutionary optimisation algorithms are also developed on the basis of nature and social inspired ideas. Optimisation forms an essential part of the modelling framework and is employed in the direct optimal design problems as well. The proposed cooperative fuzzy modelling methodology and the devised evolutionary optimisation algorithms are then applied to knowledge discovery in terms of systems modelling and control (static optimisation via reverse-engineering), using simulation platforms as well as real industrial data.
The experimental results show that the proposed modelling framework and optimisation algorithms outperform some of the other salient techniques; the proposed approaches can successfully work within the context of the high-dimensional industrial applications, including modelling and optimal design problems.
