| Summary: | 碩士 === 樹德科技大學 === 資訊管理系碩士班 === 99 === Small data set problems often make it difficult for researchers to obtain robust conclusions. However, such problems often arise, such as in the prediction of a terrorist attacks, new disease diagnoses, and pilot product management knowledge. Essentially, in such cases, either insufficient samples or prediction attributes mean it is statistically hard to build reliable models for extracting useful knowledge.
Faced with a specific small data set problem, this research first applies extreme value theory to estimate the domain range of an attribute, and then uses a virtual sample generation technique to fill the information gaps in the sparse data. Further, the sinusoid combination of real attributes is developed to construct a virtual attribute, which is used to find a more effective attribute to build the knowledge model. A real data set is employed to validate the performance of the proposed method by comparing it with the Artificial Neural Network and Mega-Trend-Diffusion approaches. The results indicate that the prediction error rate can be significantly decreased by applying the proposed method to very small data sets.
|
|---|
