Classifier combinations are effective techniques
for difficult pattern recognition problems such as speech recognition
where the combination of differently trained classifiers can produce
a more robust phoneme classification on noisy datasets. In this
paper we investigate traditional linear combination schemes (e.g.
arithmetic mean and least squares methods), and propose a new combiner
based on the Analytic Hierarchy Process (AHP), a method frequently
applied in mathematical psychology and multi-criteria decision making.
In addition, we experimentally compare the applicability of these
linear combination schemes using neural network classifiers on a
speech recognition framework and two test sets from the UCI repository.
