Sea Water Pollution Assessment Based on Ensemble of Classifiers

This paper studies the problem of evaluating pollution for a number of water quality variables, based on measurements from under-water sensors. It is a common strategy in water quality assessment to attempt to select the most accurate classifier to predict the class of the particular test samples. This approach reflects the reasonable belief that enhanced classifier of basic dataset can be used to improve assessment modeling. However, nature is complex, and even the most detailed classification model is extremely simple in comparison. At some point, additional detail exceeds our ability to assess and predict water pollution with reasonable error levels. In those situations, an attractive alternative may be to dynamically select most suitable set of classifiers to label the test samples. This viewpoint is the basis for consideration of selection of classifier ensemble for water pollution assessment and prediction. It is shown theoretically and experimentally that choosing the set of classifiers, from a population of high accurate classifiers, with lowest inaccuracy among its members leads to increase the level of confidence of classification, consequently, increasing the generalization performance. Experimental results provide interesting insights on the predictability of the water quality and the performance of our method comparing with support vector machine classifier.