Comparison of knowledge elicitation techniques in the domain of electronic filter tuning

The work done towards the construction of an expert system to assist an operator in the identification of the corrective action to be applied during tuning of electronic filters is described. The first part introduces two algorithms for induction by examples (ID3 and adaptive combiner) and their relationship to expert systems. The two algorithms were applied, in a series of tests which involved an incremental presentation of a number of examples, to the task of filter tuning. The reported results suggest the use of ID3 when a small number of classes is present. The second part presents subsequent work with ID3. Results are reported of using this algorithm for filter tuning with examples containing either numerical or logical attribute values. A comparison of the results showed that improved test performance was achieved by using logical values