Evolving Computation Offers Potential for Estimation of Pest Establishment

This paper introduces an evolving computational and a statistical model for quantitatively estimating the establishment potential of a pest insect and compares their performances. The models were used to predict the establishment potential of Planocuccus citri (Risso), the citrus mealybug. They have the common clustering and probability evaluation modules, but very different regression modules. The evolving computational model uses a dynamic neuro-fuzzy inference system to build the estimation function. The statistical model employs a multiple linear regression model in its final stage. The evolving computational model is preferred because (1) it creates fuzzy rules providing knowledge about how the pest responds to influential environmental variables, and (2) it is able to accept new data during its operation. The model can be incrementally trained and rules can be extracted from the model explaining the relationship between conditions and probability of pest establishment.