A Lazy Model-Based Algorithm for On-Line Classification

This paper presents a lazy model-based algorithm, named DBPredic-tor, for on-line classification tasks. The algorithm proposes a local discretization process to avoid the need for a lengthy preprocess stage. Another advantage of this approach is the ability to implement the algorithm with tightly-coupled SQL relational database queries. To test the algorithm’s performance in the presence of continuous attributes an empirical test is reported against both an eager model-based algorithm (C4.5) and a lazy instance-based algorithm (k-NN).