Use of neural networks as a tool to assess pesticide contamination potential of rural domestic wells

Artificial Neural Networks were used to assess nitrate and pesticide contamination potential of rural private wells using data from 192 drilled and driven wells and 163 large-diameter dug and bored wells. Four separate models, two for the two well types and one each for pesticide and nitrate, were developed for training and testing. Input parameters were fed as discrete values belonging to various ranges of values (like that represented in a histogram format) and the output was a set of four values with high, medium, low, or no impacts. While the training efficiency of the network reached between 95 and 100 percent for these four models, the prediction accuracy for the four models ranged from a low of slightly above 50 percent for nitrate in dug and bored wells to a high of 90 percent for pesticides in drilled and driven wells. The initial results of this study indicates that ANN may be potential screening tool to assess well contamination.