Attribute Selection Method And Its Application Combined With Mapreduce-based Improved BACO And Fractal Dimension

Attribute selection is a primary preprocessing step for reducing dimension of datasets in data mining: Feature selection method focuses on reducing the noisy data and redundant attributes while keeping the core value of original datasets.So this method can effectively save storage space as well as computing resources and improve the accuracy of decision analysis.The key of feature selection problem is to find critical attributes in many feature dimensions.Therefore,each feature can be represented by two states: critical attribute(status ‘1')or non-critical attribute(status ‘0').On that basis,Binary Ant Colony optimization(BACO)is used as a searching strategy,while fractal dimension is proposed as evaluation standard of the subset for solving the attribute reduction problem.Considering the shortcomings of BACO,like consuming too much time,easily dropping into local optima,failing in parallel solving,a series of ways are proposed to improve its performance.This thesis combines the improved BACO and fractal theory in the field of attribute selection.The main research work and results are summarized as follows: Firstly,an improved Binary Ant Colony Optimization Algorithm is proposed by introducing an ariable parameter selection of location strategy,a cross variation strategy for partial optimization and a new pheromone updating rule with blocking mechanism.Secondly,the computational efficiency of current attribute selection methods is difficult to meet the processing requirements of high dimensional data under cloud computing environment,so a parallelized algorithm(MRIBACO)is proposed based on MapReduce programming model,and the Map and Reduce function of the algorithm are defined.Thirdly,MRIBACO algorithm is used as search strategy of discrete solution space combined with fractal dimension to deal with feature selection problem.The trials on the UCI dataset show the effectiveness and stability of the algorithm.Finally,the method is applied in the field of haze prediction to analyze the key factors in Beijing,Shanghai and Guangzhou.The experimental results show that the reduction results have higher credibility,which provides a theoretical basis for prevention and control of haze climate.