Mass Data Reduction Based On Rough Set And Its Applications

Mass Data Reduction has always been the key issues and research topic in the area ofdata mining, machine learning and pattern recognition. For this reason, on the basis of theanaylsis of the commonly rough set attribute reduction algorithm and the attribute reductionalgorithm based on the Evolutional Computation (EC), a hybrid optimization algorithm(QPSO-DE) is presented based on the quantum particle swarm optimization (QPSO) andDifferential Evolution (DE), then the mass data reduction method based on QPSO-DE hybridalgorithm is proposed, which is verified by simulation with the typical examples and used inactual application. The main works are as follows:(1)Analysis and simulation on commonly attribute reduction algorithms. The attributereduction algorithms based on discernibility matrix and attribute significance are simulatedwith the typical examples. The drawbacks of commonly attribute reduction algorithms arepointed out that attribute reduction task can not be fulfilled with the current algorithms whenInformation Table’s data dimension is larger than10000and the complexity of the algorithmis increased exponentially.(2)Study on the QPSO-DE hybrid algorithm. Drawbacks of the PSO,QPSO,DE is alsopointed out that when the optimizational problem is multi-modal the algorithm will be trappedinto the local optimal due to the monotonity of the information sharing mechanism and theswarm’s diversity. For this reason, the QPSO and DE algorithm are merged into a new hybridoptimizational algorithm. Its superiority over the current algorithms is indicated by thesufficient statistical experiments.(3) The method of the mass data reduction based on the QPSO-DE algorithm is presented and used in actual application. The model of the mass data reduction and fitnessfunction is constructed. The reductive data subset is computed using the QPSO-DE attributereduction algorithm. Its superiority is verified by the simulation with the typical examples.Then it is applied to the process of the logging information. An Oil-layer recognition systembased on Computational Intelligence is designed. It indicated through actual application thatnot only is the reductive subset computed quickly, but also the system model construction andrecogonitional task can be carried out efficiently by using the proposed algorithm.