Abstract
Existing approaches to the inductive learning problem include Symbolic and Connectionist algorithms. While the Symbolic approach is generally found to run significantly faster during learning, the Connectionist algorithms are often more accurate at classifying novel examples in the presence of noisy data. This paper presents a technique that determines the topology and initial weightsof a neural network using a decision tree, thus combining both approaches. Experimental results on benchmark real-world datasets indicate that this technique outperforms the above mentioned approaches both in efficiency and accuracy.