ARTIFICIAL KEYS FOR BOTANICAL IDENTIFICATION USING A MULTILAYER PERCEPTRON NEURAL-NETWORK (MLP)

In this paper, practical generation of identification keys for biologi cal taxa using a multilayer perceptron neural network is described. Un like conventional expert systems, this method does not require an expe rt for key generation, but is merely based on recordings of observed c haracter states. Like a human taxonomist, its judgement is based on ex perience, and it is therefore capable of generalized identification of taxa. An initial study invoking identification of three species of Ir is with greater than 90% confidence is presented here. In addition, th e horticulturally significant genus Lithops (Aizoaceae/Mesembryanthema ceae), popular with enthusiasts of succulent plants, is used as a more practical example, because of the difficulty of generation of a conve ntional key to species, and the existence of a relatively recent monog raph. It is demonstrated that such an Artificial Neural Network Key (A NNKEY) can identify more than half (52.9%) of the species in this genu s, after training with representative data, even though data for one c haracter is completely missing.