ALBERT

Description: The classification of stream waters using parameters such as fecal coliforms into the classes of body contact and recreation, fishing and boating, domestic utilization, and danger itself is a significant practical problem of water quality prediction worldwide. Various statistical and causal approaches are used routinely to solve the problem from a causal modeling perspective. However, a transparent process in the form of Decision Trees is used to shed more light on the structure of input variables such as climate and land use in predicting the stream water quality in the current paper. The Decision Tree algorithms such as classification and regression tree (CART), iterative dichotomiser (ID3), random forest (RF), and ensemble methods such as bagging and boosting are applied to predict and classify the unknown stream water quality behavior from the input variables. The variants of bagging and boosting have also been looked at for more effective modeling results. Although the Random Forest, Gradient Boosting, and Extremely Randomized Tree models have been found to yield consistent classification results, DTs with Adaptive Boosting and Bagging gave the best testing accuracies out of all the attempted modeling approaches for the classification of Fecal Coliforms in the Upper Green River watershed, Kentucky, USA. Separately, a discussion of the Decision Support System (DSS) that uses Decision Tree Classifier (DTC) is provided.