A Lazy One-Dependence Classification Algorithm Based on Selective Patterns

Data mining is a widely acceptable method on mining knowledge from large databases, and classification is an important technique in this research field. A naive Bayesian classifier is a simple but effective probabilistic classifier, which has been widely used in classification. It is commonly thought to assume that the probability of each attribute belonging to a given class value is independent of all other attributes in the naive Bayesian classifier; however, there are lots of contexts where the dependencies between attributes are complex and should thus be considered carefully. It is an important technique that constructing a classifier using specific patterns based on "attribute-value" pairs in lots of researchers' work, and the classification result will be impacted by dependencies between these specific patterns meanwhile. In this paper, a lazy one-dependence classification algorithm based on selective patterns is proposed, which utilizes both the patterns' discrimination and dependencies between attributes. The classification accuracy benefits from mining and employing patterns which own high discrimination, and building the one-dependence relationship between attributes in a proper way. Through an exhaustive experimental evaluation, it shows that the proposed algorithm is competitive in accuracy with the state-of-the-art classification techniques on datasets from the UCI repository.