In this paper the allozyme pattern recognition algorithm was designed and implemented for identification of Rhagoletis pomonella species complex. The technique of data mining was used to search the allozyme dataset for the pattern association rules and estimate the probability for each phenotype on nine different taxa and seventeen loci. The pattern recognition models were built based on information of host fidelity of R. pomonella species group and phenotype frequencies and associations on loci. The controller of the algorithm is called pattern recognition organizer. The organizer controls and integrates the functioning of logic engine and Bayesian classifier. The system receives a new allozyme pattern and sends taxa with some membership that links to the pattern. By using 3,888 allozyme samples provided by Dr. Berlocher to train and test the system, the results show that the average success rate of the classification system is 79.1% with performance confidence interval [0.767, 0.813]. Among nine taxa R. zephyria and R. cornivora can be identified without error (100% success rate). R. mendax, and sparkleberry fly are also classified with very high success rates of 89% and 87% respectively. Classification performance of R. pomonella apple race is the lowest with success rate of 58%. The mayhaw fly and plum fly are not on the testing dataset due to their small sample size.
Species 1: Diptera Tephritidae Rhagoletis pomonella (apple maggot)
Species 2: Diptera Tephritidae Rhagoletis cornivora
Species 3: Diptera Tephritidae Rhagoletis mendax
Keywords: allozyme, pattern recognition
The ESA 2001 Annual Meeting - 2001: An Entomological Odyssey of ESA