Classification, Association and Clustering of Water Body Data: Application to Water Quality Monitoring

Abstract

The application of advanced computational resources in the support of environmental management systems has increased in recent decades. The objective of this research is to discover useful knowledge amidst water monitoring data collected between 2005 and 2011, in a specific region of the state of Sao Paulo in Brazil. The research was performed in three steps: discovery of classification rules of ecotoxicity in water samples using predictive modelling; investigation of the presence of strong relationships between water quality parameters by associative analysis; and discovery of sampling sites, which are similar with respect to their quality parameters, by applying a clustering method. The knowledge discovery process employed in this work was divided into four phases: selection of study area and data, pre-processing of the selected data, mining of the pre-processed data, and interpretation and evaluation of the results. The data mining step was supported by different algorithms, each one focused on a particular issue of the research: sequential coverage algorithm for classification of ecotoxicity; Apriori algorithm for identifying associations between quality parameters, and Prim’s algorithm and pruning algorithm for clustering of water sampling sites. We observed the association of certain quality parameters to water ecotoxicity, the existence of correlations between some of the quality parameters, and the presence of homogeneous groups amidst the sampling sites. These results may provide valuable input to the field of water quality monitoring, from the technical activities, such as the laboratorial analysis of water samples, to the decision makers responsible for defining the future public policies for water resources management.