Knowledge of water quality is an important requirement in planning, developing, and protecting water resources. Therefore, it is essential to determine the quality of water for various uses, including irrigation in different areas. The aim of this study is to evaluate the performance of different data mining methods using water quality parameters measured in Aladag River. At the first stage, the water quality in the Aladag River in Turkey was classified by United States Salinity Laboratory (USSL) diagrams and the Schofield 1933, Schofield 1935, and Wilcox-Magistad methods. Support vector classifier (four different kernels), k-nearest neighbour, and decision tree methods-including Hoeffding tree, random forest, random tree, and REP tree methods were used to classify the water quality. Hydro-chemical and hydrological parameters were used in the analysis, based on alternative scenarios. The methods were compared for accuracy based on kappa statistics and error rates. Evaluation of the data mining methods reflected the high accuracy and performance of these methods in classifying the water quality. In general, from the kernels used for the support vector classifier method, the Pearson universal kernel and from the decision trees, the REP tree provided the best results for Aladag River water quality classification. For the k-nearest neighbour method, the best results were obtained using the four nearest neighbours and Euclidean distances.
展开▼
