Parallel learning using decision trees: a novel approach

Decision trees are one of the most effective and widely used induction methods that have received a great deal of attention over the past twenty years. When decision tree induction algorithms used with uncertain rather than deterministic data, the result is complete tree, which can classify most of the unseen samples correctly. This tree would be pruned in order to reduce its classification error and over-fitting. Recently, parallel decision tree researches concentrated on dealing with large databases in reasonable amount of time. In this paper we present new parallel learning methods that are able to induce a decision tree from some overlapping partitioned training set. Our methods are based on combination of multiple induction methods; each one is running on different processor. These methods have been developed based on Kramer and fuzzy mode to control and combine the result of learning methods in order to generate the final tree. Experimental results show that if the attributes and classes in training set have uniform distribution and the size of training set are not too small, these methods result statistically lower error rate in comparison to existing methods.