| Summary: | 碩士 === 國立成功大學 === 工業與資訊管理學系 === 104 === As the global competition getting more and more intense, it also leads to the shorter product life cycle. Reducing the time and cost of pilot-run can enhance the competitive ability of enterprises effectively, somehow the small dataset learning problems will also occur as the same time. There exists no appropriate statistics tool to evaluate the population when the sample size is too small, but we can fix the problem through virtual sample generation methods, which is widely used in industry and machine learning area. There are very few studies deal with nominal attributes due to the limit on domain estimation methods, therefore, this paper proposes a method that generate virtual sample based on the discrete degree of nominal attributes, then estimate the general population domain by fuzzy membership function. Two learning models will be used to test the efficiency of proposed method, including backpropagation neural network and support vector regression, and then the Wilcoxon-sign test will be used to test the difference with raw dataset. The result shows that the propose method can reduce the mean absolute error (MAE) as well as enhance classification accuracy by generating nominal virtual samples.
|
|---|
