Diagnostic Reliability Maximization for Weighted Voting Classification System Using Swarm-based Optimization Algorithm

Abstract

The weighted voting classification system consists of n units (or individuals) from beginner to expert, each of which provides a binary decision (0 or 1), or abstains from voting. The system output is 1 if the cumulative weight of all 1-opting units is at least a pre-specified fraction t of the cumulative weight of all non-abstaining units. Otherwise the system output is 0. However, to evaluate the reliability of a weighted voting classification system (WVCS) is very difficult because of its combinatorial complexity from different unit weights or dynamic threshold value. Hence, a method is suggested which allows the reliability of weighted voting system to be exactly evaluated. The approach is based on using a universal generating function methodology. Then, a swarm-based optimization algorithm is implemented to explore proper unit weight and threshold value for the maximization of diagnostic reliability. A particle swarm optimization (PSO) is used as the optimization tool. This work adopts the breast cancer dataset in UCI Machine Learning Repository to verify classification effectiveness of WVCS. In terms of ten-fold cross-validation, the comparative study is made between an optimized WVCS and the other approaches.