Comparison of ID3 and its generalized version

A generalized ID3 learning algorithm, which possesses more processing capabilities than the original ID3, is proposed. This generalized ID3 learning algorithm can manage uncertain training instances and consider the different importance of different training instances in finding an appropriate decision tree in a noisy environment. If appropriate priori domain knowledge is available, this algorithm can further utilize it in reducing the effect of noise. Relations on generality, accuracy, tree complexity and time complexity of ID3 and generalized ID3 are also discussed, with experiments verifying their correctness