G. Štiglic, M. Lenič, and P. Kokol (Slovenia)
feature selection, rough sets, machine learning
The problem of feature selection is an important part of data processing prior to applying a learning algorithm. By adding only the most relevant features to the subset, we can improve the results of the learning algorithm. There are two common approaches: a wrapper uses learning algorithm to evaluate the relevance of selected features, while a filter evaluates features according to heuristics based on characteristics of the data. In our proposed method we search for the most relevant features from the subset of features that are usually left out by classic rough sets approach. We compare our wrapper approach to classical decision trees method. The feature subsets selected by our method are much smaller than subsets used in decision trees approach.
Important Links:
Go Back
