Attributes reduction model with user preferences

In rough set theory, Attributes Reduction algorithms are utilized to extract patterns or rules from the table-formed decision systems. The attributes reduction algorithms aim to find the minimum subset of attributes which can distinguish all the items of different classes. However, most existing attributes reduction algorithms over-focus the distinguishability of attributes and neglect the user requirements in real applications. To tackle this problem, we propose an attributes reduction model with preferences to discover patterns according to user interests. In the proposed model, user preferences in data mining task are formally represented by attribute orders. For the attributes reduction algorithm implementation, we design the novel data structure of linked list to storage the elements for discerning pairwise items and adopt discernibility thresholds to avoid overfitting. The proposed attributes reduction model with user preferences is performed on UCI data sets. Experimental results demonstrate that the proposed attributes reduction model is effective to extract the data patterns consistent with user requirements.