Identification of Africanized honeybees via nonlinear multilayer perceptrons

Africanized (“killer”) honeybees represent an immediate and serious threat to public and agricultural well-being in the southern United States due to their recent immigration from Central and South America. Discrimination of hybrid Africanized bees from native European-stock bees is problematic, and the current linear statistical tools used for this purpose are not totally reliable. The authors describe the use of multilayer perceptron neural networks for classifying bees on the basis of quantitative wing-vein traits and evaluate their performance with respect to linear discriminant functions. The networks generalize significantly better than linear functions, with success rates consistently greater than 95%. Thus these preliminary results are very encouraging and promise a powerful approach to the identification of hybrid Africanized honeybees