Published online by Cambridge University Press: 12 December 2011
The oriental fruit fly Bactrocera dorsalis (Hendel) is a very serious pest of fruit trees, causing enormous economic losses globally. The present study examines the capability of an artificial neural network (ANN) with a Quasi-Newton (QN) algorithm to predict a fruit fly trap catch and compare the results with those of a traditional regression model. MATLAB 7.0 was used to develop ANN programming and the fortnightly measurement of 14 input variables (abiotic along with biotic variables) provided the database for analysing the ANN model. An input model using a total of 14 identified input nodes with a selected QN-ANN structure (14-25-20-1) gave an optimum result. In general, the present study showed that an ANN could be used to estimate fruit fly trap catch with enhanced accuracy (R2 = 0.92; root mean square error (RMSE) = 23.75; Nash–Sutcliffe efficiencies = 0.99) over traditional regression models (R2 = 0.76; RMSE = 30.28; Nash–Sutcliffe efficiencies = 0.76). This finding helps the region-specific fruit fly monitoring and management programmes that lack long-term historic data.