Abstract
Data mining is the art of discovering knowledge from databases. The user specifies the type of patterns to be mined, and the miner uses techniques to find the required patterns. Many techniques have been introduced for mining traditional patterns like frequent itemsets, association rules, etc. However, mining patterns with additional properties remains a bottleneck for specialists nowadays due to the algorithmic effort needed to handle these properties.Recently, researchers have taken advantage of the flexibility of constraint programming to model various data mining problems. In terms of CPU time, constraint programming-based methods have not yet competed with ad hoc algorithms. However, their flexibility allows the modeling of complex user queries without revising the solving process.In this thesis we propose to use constraint programming for modeling and solving some well known data mining problems.Our first contribution is a constraint programming model for mining association rules. To implement our model, we introduce a new global constraint, CONFIDENT, for ensuring the confidence of rules.We prove that completely propagating CONFIDENT is NP-hard. We thus provide a non-complete propagator and a decomposition for CONFIDENT. We also capture the minimal non-redundant rules, a condensed representation of association rules, by introducing the global constraint GENERATOR. GENERATOR is used for mining itemsets that are generators. For this constraint, we propose a complete polynomial propagator.Our second contribution is a generic framework based on constraint programming to mine both borders of frequent itemsets, i.e. the positive border or maximal frequent itemsets and the negative border or minimal infrequent itemsets. One can easily decide which border to mine by setting a simple parameter. For this, we introduce two new global constraints, FREQUENTSUBS and INFREQUENTSUPERS, with complete polynomial propagators. We then consider the problem of mining borders with additional constraints. We prove that this problem is coNP-hard, ruling out the hope for the existence of a single CSP solving this problem (unless coNP is in NP).