Synthesis and Optimization of Data Mining Models to Achieve Higher Performance

点击翻译按钮获取中文摘要

We have studied the synthesis and optimization of mathematical models for data mining to achieve higher performance, and obtained the following results :1.A new stochastic optimization algorithm for discrete optimization :Based on stochastic sensitivity analysis techniques, a novel optimization algorithm using Gaussian white noise is developed for a class of discrete optimization problems. Unlike the familiar local search algorithms, the proposed method does not require time-consuming metaheuristics. The algorithm generates sample points with Gaussian white noise and, and yields gradient information that are needed in local search. By intensive numerical studies, it is shown that the method is effective in finding solutions for noisy objective landscapes with inexact gradient information including traveling salesman problem (TSP).2. Synthesis of data mining models to achieve high performance :A technical framework is developed to assess the performance of data mining models and synthesize them in order to construct a mixture of experts. Based on the boosting techniques, the proposed framework enables synthesis of the near-optimal mixture. The numerical study and analysis are conducted on real business data and it is validated that the framework is effective in practice.3.Applications to production scheduling and knowledge management :Applications of data mining are studied to develop production scheduling and knowledge management. The aspect of knowledge discovery is especially explored to build soft systems methodology in order to incorporate knowledge-creation theory.