שמעון ליאור רוקח

Data mining algorithms aim at searching interesting patterns in large amount of data in manageable complexity and good accuracy. Decomposition methods are used to improve both criteria. As opposed to most decomposition methods, that partition the dataset via sampling, this paper presents an accuracyoriented method that partitions the instance space into mutually exclusive subsets using K-means clustering algorithm. After employing the basic divide-andinduce method on several datasets with different classifiers, its error rate is compared to that of the basic learning algorithm. An analysis of the results shows that the proposed method is well suited for datasets of numeric input attributes and that its performance is influenced by the dataset size and its homogeneity. Finally, a homogeneity threshold is developed, that can be used for deciding whether to decompose the data set or not.