A case study using support vector machines, neural networks and logistic regression in a GIS to identify wells contaminated with nitrate-N

摘要: Accurate and inexpensive identification of potentially contaminated wells is critical for water resources protection management. The objectives this study are to 1) assess the suitability approximation tools such as neural networks (NN) support vector machines (SVM) integrated in a geographic information system (GIS) identifying 2) use logistic regression feature selection methods identify significant variables transporting contaminants through soil profile groundwater. Fourteen GIS derived hydrogeologic landuse parameters were used initial inputs study. Well quality data (nitrate-N) from 6,917 provided by Florida Department Environmental Protection (USA) an output target class. reduced number input nine. Receiver operating characteristics (ROC) curves evaluation these tools. Results showed superior performance with NN compared SVM especially on training while testing results comparable. Feature did not improve accuracy; however, it helped increase sensitivity or true positive rate (TPR). Thus, higher TPR was obtainable fewer variables.