Selecting Features for Impact Identification using Fuzzy Clustering

Q. Shan and G. King (UK)


Signal Processing, Fuzzy C-Means Clustering, Feature Selection, Impact Identification.


This paper presents a method of signal preprocessing and feature selection for impact identification. Original impact signals are preprocessed with a cross correlation algorithm. The correlation results are processed with a fuzzy c-means clustering algorithm for selecting data points. The selected data points in each cluster are computed by the method of taking means. The resulting average values are the final features for impact identification. The mechanism of feature extraction addressed in this study aims to reduce the dimensionality of training data and the input space, enhancing the accuracy of impact identification, and reducing the complexity of the architecture. In addition the ANFIS adaptability is improved, thus avoiding overfitting.

Important Links:

Go Back