Backfilling missing microbial concentrations in a riverine database using artificial neural networks.

Predicting peak pathogen loadings can provide a basis for watershed and water treatment plant management decisions that can minimize microbial risk to the public from contact or ingestion. Artificial neural network models (ANN) have been successfully applied to the complex problem of predicting peak pathogen loadings in surface waters. However, these data-driven models require substantial, multiparameter databases upon which to train, and missing input values for pathogen indicators must often be estimated. In this study, ANN models were evaluated for backfilling values for individual observations of indicator bacterial concentrations in a river from 44 other related physical, chemical, and bacteriological data contained in a multi-year database. The ANN modeling approach provided slightly superior predictions of actual microbial concentrations when compared to conventional imputation and multiple linear regression models. The ANN model provided excellent classification of 300 randomly selected, individual data observations into two defined ranges for fecal coliform concentrations with 97% overall accuracy. The application of the relative strength effect (RSE) concept for selection of input variables for ANN modeling and an approach for identifying anomalous data observations utilizing cross validation with ANN model are also presented.

Artificial neural network prediction of viruses in shellfish.

A database was probed with artificial neural network (ANN) and multivariate logistic regression (MLR) models to investigate the efficacy of predicting PCR-identified human adenovirus (ADV), Norwalk-like virus (NLV), and enterovirus (EV) presence or absence in shellfish harvested from diverse countries in Europe (Spain, Sweden, Greece, and the United Kingdom). The relative importance of numerical and heuristic input variables to the ANN model for each country and for the combined data was analyzed with a newly defined relative strength effect, which illuminated the importance of bacteriophages as potential viral indicators. The results of this analysis showed that ANN models predicted all types of viral presence and absence in shellfish with better precision than MLR models for a multicountry database. For overall presence/absence classification accuracy, ANN modeling had a performance rate of 95.9%, 98.9%, and 95.7% versus 60.5%, 75.0%, and 64.6% for the MLR for ADV, NLV, and EV, respectively. The selectivity (prediction of viral negatives) was greater than the sensitivity (prediction of viral positives) for both models and with all virus types, with the ANN model performing with greater sensitivity than the MLR. ANN models were able to illuminate site-specific relationships between microbial indicators chosen as model inputs and human virus presence. A validation study on ADV demonstrated that the MLR and ANN models differed in sensitivity and selectivity, with the ANN model correctly identifying ADV presence with greater precision.

Probing Norwalk-like virus presence in shellfish, using artificial neural networks.

A database was examined using artificial neural network (ANN) models to investigate the efficacy of predicting PCR-identified Norwalk-like virus presence and absence in shellfish. The relative importance of variables in the model and the predictive power obtained by application of ANN modelling methods were compared with previously developed logistic regression models. In addition, two country-specific datasets were analysed separately with ANN models to determine if the relative importance of the input variables was similar for geographically diverse regions. The results of this analysis found that ANN models predicted Norwalk-like virus presence and absence in shellfish with equivalent, and better, precision than logistic regression models. For overall classification performance, ANN modelling had a rate of 93%, vs 75% for the logistic regression. ANN models were able to illuminate the site-specific relationships between indicators and pathogens.

A neural-network-based classification scheme for sorting sources and ages of fecal contamination in water.

Artificial neural networks (ANNs) were successfully applied to data observations from a small watershed consisting of commonly measured indicator bacteria, weather conditions, and turbidity to distinguish between human sewage and animal-impacted runoff, fresh runoff from aged, and agricultural land-use-associated fresh runoff from that of suburban land-use-associated-fresh runoff. The ANNs were applied in a cascading, or hierarchical scheme. ANN performance was measured in two ways: (1) training and (2) testing. An ANN was able to sort sewage from runoff with < 1% error. Turbidity was found to be relatively unimportant for sorting sewage from runoff, while gross measurements of gram-negative and gram-positive bacteria were required. Predictions clustered tightly around the known values. ANN classification of aged suburban runoff from fresh, and agricultural runoff from suburban was accomplished with > 90% accuracy.

Neural network modelling of Cryptosporidium and Giardia concentrations in the Delaware River, USA.

Artificial neural networks are brain-like structures used in mathematical modelling that excel in pattern recognition. In this research, a simple feed-forward artificial neural network, trained by error back-propagation algorithm, was used as a tool to relate peak Cryptosporidium and Giardia concentrations with other biological, chemical and physical parameters in surface water. Multiple water quality parameters at a water treatment plant intake on the Delaware River, New Jersey, USA, collected in 1996, were provided to the authors for recognition analysis. Water samples were classified as "background" and "above background" based on the concentration of full and empty oocysts and cysts of Cryptosporidium and Giardia. The results of this preliminary effort were encouraging. Parameters significant to the identification of each protozoa were identified, eight for Cryptosporidium and seven for Giardia by a stepwise elimination technique. Data withheld from the model training was used to validate the trained models and evaluate the most effective internal architecture. In both cases, the best prediction performance was found when the number of internal nodes was twice that of the input parameters in single hidden-layer architecture. Predictions for the classification of the verification data set resulted in no false-negatives (mis-prediction of above background protozoa concentrations) when the models were optimally trained.