A fuzzy-based instance selection approach for data mining

Data mining is an area bat is enjoying increasing growth. One of the most time-consuming tasks in data mining is data preparation or preprocessing. Because data preprocessing Cakes more time and effort than the rest of the data mining process [13], the need for improved data preprocessing methods is well recognized [4; 8]. Dealing with missing values can further complicate data preprocessing. Several methods have been used to resolve the problem of instance selection when there are missing data values. Most of these methods 1) discard records with missing values; 2) use all records and ignore missing values; or 3) use all records and infer missing values. These methods do not consider the utility of individual attributes. Here we introduce a fuzzy-based information metric that considers the usefulness of the individual attributes by incorporating domain knowledge into a multi-criteria decision-making instance selection technique.