Pawel Terlecki

On the relation between rough set reducts and jumping emerging patterns

Abstract This paper presents the relations between rough set reducts and jumping emerging patterns. Observations are introduced formally and supported by brief examples. Furthermore, we propose practical applications of these observations to the minimal reduct problem and to testing whether a given attribute set is differentiating. We believe that our study can be expanded so as to include other types of reducts and emerging patterns.

Efficient Discovery of Top-K Minimal Jumping Emerging Patterns

Jumping emerging patterns, like other discriminative patterns, help to understand differences between decision classes and build accurate classifiers. Since their discovery is usually time-consuming and pruning with minimum support may require several adjustments, we consider the problem of finding top-kminimal jumping emerging patterns. We describe the approach based on a CP-Tree that gradually raises minimum support during mining. Also, a general strategy for pruning non-minimal patterns and their descendants is proposed. We employ the concept of attribute set dependence to test pattern minimality. A two and multiple class version of the problem is discussed. Experiments evaluate pruning capabilities and execution time.

Jumping emerging patterns with negation in transaction databases - Classification and discovery

This paper examines jumping emerging patterns with negation (JEPNs), i.e. JEPs that can contain negated items. We analyze the basic relations between these patterns and classical JEPs in transaction databases and local reducts from the rough set theory. JEPNs provide an interesting type of knowledge and can be successfully used for classification purposes. By analogy to JEP-Classifier, we consider negJEP-Classifier and JEPN-Classifier and compare their accuracy. The results are contrasted with changes in rule set complexity. In connection with the problem of JEPN discovery, JEP-Producer and rough set methods are examined.

Jumping Emerging Pattern Induction by Means of Graph Coloring and Local Reducts in Transaction Databases

This paper demonstrates how to employ rough set framework in order to induce JEPs in transactional data. The algorithm employs local reducts in order to generate desired JEPs and additional EPs. The number of the latter is decreased by preceding reduct computation with item aggregation. The preprocessing is reduced to graph coloring and solved with efficient classical heuristics. Our approach is contrasted with JEP-Producer, the recommended method for JEP induction. Moreover, a formal apparatus for classified transactional data has been proposed.

Local projection in jumping emerging patterns discovery in transaction databases

This paper considers a rough set approach for the problem of finding minimal jumping emerging patterns (JEPs) in classified transactional datasets. The discovery is transformed into a series of transactionwise local reduct computations. In order to decrease average subproblem dimensionality, we introduce local projection of a database. The novel algorithm is compared to the table condensation method and JEP-Producer for sparse and dense, originally relational data. For a more complete picture, in our experiments, different implementations of basic structures are considered.

Local reducts and jumping emerging patterns in relational databases

This paper refers to the notion of minimal pattern in relational databases. We study the analogy between two concepts: a local reduct, from the rough set theory, and a jumping emerging pattern, originally defined for transactional data. Their equivalence within a positive region and similarities between eager and lazy classification methods based on both ideas are demonstrated. Since pattern discovery approaches vary significantly, efficiency tests have been performed in order to decide, which solution provides a better tool for the analysis of real relational datasets.

Adaptive classification with jumping emerging patterns

In this paper a generic adaptive classification scheme based on a classifier with reject option is proposed. A testing set is considered iteratively, accepted, semi-labeled cases are used to modify the underlying hypothesis and improve its accuracy for rejected ones. We apply our approach to classification with jumping emerging patterns (JEPs). Two adaptive versions of JEP-Classifier, by support adjustment and by border recomputation, are discussed. An adaptation condition is formulated after distance and ambiguity rejection strategies for probabilistic classifiers. The behavior of the method is tested against real-life datasets.

Attribute set dependence in reduct computation

In the paper we propose a novel approach to finding roughset reducts in information systems. Our method combines an apriorilikescheme of space traversing with an efficient pruning condition basedon attribute set dependence. Moreover, we discuss theoretical and implementationalaspects of our pruning procedure, including adopting abst and a trie tree for storing set collections. Operation number andexecution time tests have been performed in order to demonstrate theefficiency of our approach.

Local Table Condensation in Rough Set Approach for Jumping Emerging Pattern Induction

This paper extends the rough set approach for JEP induction based on the notion of a condensed decision table. The original transaction database is transformed to a relational form and patterns are induced by means of local reducts. The transformation employs an item aggregation obtained by coloring a graph that re0ects con0icts among items. For e±ciency reasons we propose to perform this preprocessing locally, i.e. at the transaction level, to achieve a higher dimensionality gain. Special maintenance strategy is also used to avoid graph rebuilds. Both global and local approach have been tested and discussed for dense and synthetically generated sparse datasets.