PREDICTING WELL WATER VULNERABILITY TO CARBON TETRACHLORIDE CONTAMINATION USING DECISION TREE AND NEURAL NETWORK CLASSIFICATION ALGORITHMS

Data mining was performed on a dataset of VOC concentrations in private and public well water supplies with associated attribute information to examine trends in factors explaining the vulnerability of well water to VOC contamination. Aquifer vulnerability relies on the presence of a contamination source, the geochemical setting controlling contaminant concentrations, and the ability with which water and contaminants can travel through an aquifer, as well as on the location, design, construction, operation, and maintenance of the well. In this study, attributes were analyzed that affect the vulnerability of well water to contamination for the public, domestic, and irrigation wells in the dataset. Decision tree induction and a neural network approach were the machine learning techniques used for this analysis. The parameters that affect fate and transport of contamination were used as the attributes for training the classification algorithms. Based on these, the resulting models predicted either occurrence or non-occurrence of VOC contamination in the wells of the test set. The decision tree and neural network models correctly predicted the contamination class with accuracies ranging from approximately 70% to 90% for wells in the test set among different runs. Whether a well is used for domestic purposes or irrigation versus public supply was the variable which most frequently best explained the observed occurrences of VOC contamination in the well dataset; also explanatory were dissolved oxygen concentration and distance to the contamination source. The potential for developing a predictive tool based on the results was also explored. Such a tool would allow for the development of an assessment methodology useful during environmental characterization given field constraints. The methodology would prioritize the sampling of wells for carbon tetrachloride based on the prediction of vulnerability to contamination from known conditions.