A Data Mining Tool for Water Uses Classification Based on Multiple Classifier Systems

Water is not only vital for ecosystems, wildlife, and human consumption, but also for activities such as agriculture, agro-industry, and fishing, among others. However, in the same way as their water use has increased, it has also been detected an accelerated deterioration of its quality. In this sense, to have predictive knowledge about water quality conditions, can provide a significant relevance to many socio-economic sectors. In this paper, we present an approach to predict the water quality for different uses (aquaculture, irrigation, and human consumption) discovering knowledge from several datasets of American and Andean Watersheds. This proposal is based on Multiple Classifier Systems (MCS), including Bagging, Stacking, and Random Forest. Models as Naive Bayes, KNN, C4.5, and Multilayer Perceptron are combined to increase the accuracy of the classification task. The experimental results obtained show that Random Forest and Stacking expose acceptable precision on different water-use datasets. However, Bagging with C4.5 was the most appropriate architecture for the problem addressed. These results indicate that MCS techniques can be used for improving accuracy and generalization capacity of the prediction tools used by stakeholder involvement in the water quality process.