Study Of A Set Of Data Mining Techniques On E-Business Data Classification

E-business is a competitive industry, whose data value is increasingly recognized by the market. Relevant statistical methods, however, provides only statistics, statistical models and hypothesis testing conclusion. Feature of data space is yet unknown. Data mining techniques, instead, review the data from the perspective of high dimensional space and information theory to give macro-and micro-structural features, and take advantage of these features to help predict unclassified data. Different data has different field characteristics. And nominal/numerical attributes provide different classification efficiency, which makes traditional technique apply on only limited situations. Optimized general classification model requires not only overcome these difficulties, but also gives characteristics of the data space and completes the goal of classification with improved predictive performance. Generally, the scale of the dataset is large, the classifier should thus disperse computational burden in order to improve the training and prediction calculation speed. And E-business data are distributed with big flow. The data needs to be effectively integrated and understood in order to promote the business developmentThis thesis analyzes existing mining techniques and research, point out the drawbacks, e.g. there are many constraints for data types, data scale and data structure; the models are not universal fitting, computing speed is slow when dataset is big. This paper describes the causes of these shortcomings and suggests improvements with a new designed model with improvement of existing technologies. Experiments are presented with the speedup performance evaluation, which compares with the existing models. The effectiveness of the designed model and its superiority are demonstrated. And the importance of criteria threshold is also clarified.