Rough set models have been widely used as a method for feature selection in fault
diagnosis. A neighbourhood rough set model can deal with both nominal and numerical features, but selecting the neighbourhood size for its application may be a challenge. In this article, the authors illustrate that using a common neighbourhood size for all features may overestimate or underestimate a feature’s dependency degree.
The neighbourhood rough set model is then modified by setting different neighbourhood sizes for different features. The modified model is applied to the fault diagnosis of slurry pump impellers. Results show that the chosen feature subsets generated by the modified neighbourhood rough set model can be physically explained by the corresponding flow patterns and can achieve higher classification accuracy than the raw feature subsets and the feature subsets generated by the original neighbourhood rough set model.
Click the bellow link to know more