The development of a neural-based biomarker forecasting tool for classifying recreational water quality

This study will develop a computational framework capable of rapidly classifying microbial surface water quality for the protection of public health. Three classification tools will be developed using multivariate linear regression (MLR), artificial neural networks (ANN), and linear vector quantization (LVQ). The MLR and ANN approaches first quantify the microbial concentration followed by classification, while the LVQ approach directly classifies the water quality. The algorithms will be applied to microbial and hydrologic data associated with the Charles River Basin using antecedent rainfall over the previous 24 and 168-hrs and lag-1 fecal coliform concentrations as explanatory variables. Preliminary results with the MLR algorithm illustrate very good classification when the observed data is below the appropriate water quality standard (truenegativerates>95). Unfortunately, the MLR classification approach does not perform as well when the observed data is greater than the standard (truepositiverates∼50). Additional studies will be focused on evaluating the architecture and performance of both the ANN and LVQ approaches for classifying the microbial water quality, and the best performing algorithm identified.