A BORDER BASED APPROACH FOR HIDING FUZZY WEIGHTED SENSITIVE ITEMSETS

Abstract

Association rule mining is an important technique in data mining. Traditional association rule discovery process deals with crisp quantitative data values. However, there are cases when the data values are not well separable into crisp boundaries. This data can be termed as fuzzy data. In such cases, a single value can have membership associated with multiple attributes or groups. Traditional association rule discovery fails to work on such data. Fuzzy association rule mining techniques are used to deal with uncertain or fuzzy data. In many real world applications, all items in the database may not be of equal significance from data mining perspective. So, in such cases weights are assigned to items to reflect their importance. Applying privacy preservation on weighted fuzzy frequent itemsets is an active area of research in data mining. In context to privacy preservation, fuzzy weighted itemsets can be categorized as sensitive itemsets and non-sensitive itemsets. Sensitive itemsets are those which are critical to the user or application and must remain hidden. Non-sensitive itemsets are those which are less critical and may not remain hidden. Some non-sensitive itemsets have high predicting capability i.e. they may be used to predict sensitive itemsets values. It is important to identify such non-sensitive itemsets and to prevent their misuse. Also, the hiding of sensitive itemsets may affect the sanitized database. In the thesis, an algorithm has been proposed to extract fuzzy weighted frequent itemsets. The proposed work also identifies the highly predictive fuzzy weighted non-sensitive itemsets and hides them in combination of sensitive itemsets to obtain well maintained sanitized database. To achieve database sanitization, border based approach for hiding fuzzy weighted itemsets has been proposed. The, work has been done using quantitative datasets. Case data has been taken from