ABSTRACT The accuracy of the learned classification rules in data mining is affected by the used learning algorithm and the availability of the whole training set in main memory during the learning process. In this paper, we propose a combination of data reduction techniques based on attributes relevancy, data abstraction, and data generalization. We also propose a hybrid classification algorithm based on decision tree and genetic algorithm. Decision tree as a greedy algorithm is to handle generalization, where each learned rule is covered by large number of examples in the training set “large-scope rules”. The genetic algorithm handles specialization in the training set, where a small number of examples cover each of the learned rules “small-scope rules”.
