Fuzzy Entropy of Classification and its Application to Biomarker Discovery

Fuzzy entropy of classification is presented along with a criterion of maximizing the first derivative of the entropy with respect to temperature to optimize the degree of fuzziness. Fuzzy entropy of classification is used to construct classification trees comprised of multivariate fuzzy rules. These systems are of great use to scientists because of their discernable mechanism of inference. By using bootstrap Latin partitions statistically significant features can be ascertained from complex data sets. The principles are demonstrated with two simple simulated data sets.