Feature reduction of flow pattern identification by interaction index

This paper addresses the feature reduction problem in the flow pattern identification. Classification is often performed in the flow pattern identification based on data extraction from all designed instruments. However, the raw data may contain a large number of features, but many features are not necessary, even resulting in more imprecise identification. The classifier also is difficult to be determined when the number of training data needed to design a classifier grows with the number of the features. Hence, a way is needed to reduce the number of the features without losing any essential information. Data fusion method is helpful to realize feature reduction. This paper presents a new method for feature reduction using interaction index-based data fusion, and compares it with some methods presented earlier in the literature. This new method has higher correctness rate and more predictable performances than the same type of other methods.