OVERCOMING THE MYOPIA OF INDUCTIVE LEARNING ALGORITHMS WITH RELIEFF

Current inductive machine learning algorithms typically use greedy sea rch with limited lookahead. This prevents them to detect significant c onditional dependencies between the attributes that describe training objects. Instead of myopic impurity functions and lookahead, we propos e to use RELIEFF an extension of RELIEF developed by Kira and Rendell [10, 11], for heuristic guidance of inductive learning algorithms. We have reimplemented Assistant, a system for top down induction of decis ion trees, using RELIEFF as an estimator of attributes at each selecti on step. The algorithm is tested on several artificial and several rea l world problems and the results are compared with some other well kno wn machine learning algorithms. Excellent results on artificial data s ets and two real world problems show the advantage of the presented ap proach to inductive learning.