Multi-class Minimax Probability Machine

This paper investigates the multi-class minimax probability machine (MPM). MPM constructs a binary classifier that provides a worst-case bound on the probability of misclassification of future data points, based on reliable estimates of means and covariance matrices of the classes from the training data points. We propose a method to adapt MPM to multi-class datasets using the one-against-all strategy. And then we introduce an optimal kernel for MPM for each specific dataset found by genetic algorithms (GA). The proposed method was evaluated on stomach cancer data. The obtained results are better and more stable than for using a single kernel.