Data Association Analysis on Critical Features Based on Different Algorithms

Similarity evaluation methods including distance measures, cosine measure metric, have been applied on various scenarios, i.e., data clustering and classification, fault diagnosis, software engineering and pattern recognition. However, different similarity methods own distinct performance, where there is often little overlap in the similarity relationships detected by different approaches. Focus on this shortcoming, a multiple similarity-based approach is successfully presented in this paper to evaluate the inherent association among different features in real-life datasets. The purposed approach was estimated using datasets collected from a production line running in the real plant. A fused weight vector was obtained relying on the procedure. Analysis result shows that different similarities occupy distinct coefficients when similarity evaluation methods are introduced to the same dataset. Since all the similarity weight coefficients are only dependent on data to be analysed, it is reasonable to infer that compared with the single similarity method, the purposed similarity may be capable of improving the stationary and robustness of the inherent association relationship.