Diagnosis Using Multiple Classifiers and Majority-Weighted Voting (WMV)

Abstract

In this chapter, we integrate a meta-learning technique, namely weighted-majority voting (WMV) , in our diagnosis system. Multiple classifiers can be leveraged in a WMV-based diagnosis system to incorporate different sets of repair suggestions and form a single set of root-cause candidates. We review the usage of artificial neural networks (ANNs) . The advantage of ANNs is its interpretation of the relationship between the syndromes and corresponding faulty components. A trained ANN model associates the output to the weighted inputs, which derives an intuition on the contribution of each input to the output. In addition, the advantage of SVMs, as described in Chap. 2, is that the solution provided by SVMs is globally optimal and unique, while ANNs suffer from multiple local minima. Both of these two methods can be rapidly trained and they are scalable to large datasets. The proposed WMV-based system uses weights to combine the repair suggestions provided by each machine in order to identify a single set of recommended repair suggestions. The proposed WMV system can leverage results from both ANNs and SVMs.