A scalable evolutionary learning classifier system for knowledge discovery in stream data mining

Data mining (DM) is the process of finding patterns and relationships in databases.The breakthrough in computer technologies triggered a massive growth in datacollected and maintained by organisations. In many applications, these data arrivecontinuously in large volumes as a sequence of instances known as a data stream.Mining these data is known as stream data mining. Due to the large amount of dataarriving in a data stream, each record is normally expected to be processed onlyonce. Moreover, this process can be carried out on different sites in the organisationsimultaneously making the problem distributed in nature. Distributed stream datamining poses many challenges to the data mining community including scalabilityand coping with changes in the underlying concept over time.In this thesis, the author hypothesizes that learning classifier systems (LCSs) - aclass of classification algorithms - have the potential to work efficiently in distributedstream data mining. LCSs are an incremental learner, and being evolutionarybased they are inherently adaptive. However, they suffer from two main drawbacksthat hinder their use as fast data mining algorithms. First, they require a largepopulation size, which slows down the processing of arriving instances. Second,they require a large number of parameter settings, some of them are very sensitiveto the nature of the learning problem. As a result, it becomes difficult to choose aright setup for totally unknown problems.The aim of this thesis is to attack these two problems in LCS, with a specific focuson UCS - a supervised evolutionary learning classifier system. UCS is chosen as ithas been tested extensively on classification tasks and it is the supervised versionof XCS, a state of the art LCS.In this thesis, the architectural design for a distributed stream data mining systemwill be first introduced. The problems that UCS should face in a distributed datastream task are confirmed through a large number of experiments with UCS andthe proposed architectural design.To overcome the problem of large population sizes, the idea of using a NeuralNetwork to represent the action in UCS is proposed. This new system - called NLCS{ was validated experimentally using a small fixed population size and has showna large reduction in the population size needed to learn the underlying concept inthe data.An adaptive version of NLCS called ANCS is then introduced. The adaptive versiondynamically controls the population size of NLCS. A comprehensive analysis of thebehaviour of ANCS revealed interesting patterns in the behaviour of the parameters,which motivated an ensemble version of the algorithm with 9 nodes, each using adifferent parameter setting. In total they cover all patterns of behaviour noticed inthe system. A voting gate is used for the ensemble. The resultant ensemble doesnot require any parameter setting, and showed better performance on all datasetstested.The thesis concludes with testing the ANCS system in the architectural design fordistributed environments proposed earlier.The contributions of the thesis are: (1) reducing the UCS population size by anorder of magnitude using a neural representation; (2) introducing a mechanismfor adapting the population size; (3) proposing an ensemble method that does notrequire parameter setting; and primarily (4) showing that the proposed LCS canwork efficiently for distributed stream data mining tasks