Feasibility Study of Data Mining Methods Application to Estimate Aji Chai River's Water Quality Classification

Accessing to clean and high quality water sources has always been one of the main concerns of humanity. Therefore, determination of water quality is essential for various applications e.g. irrigation. In this study, irrigation water of Ajichai River in stations (Akhula, Arzanagh, Markid and Vaniar) is initially classified using USSL diagram. After that, feasibility of support vector classifier, K-nearest neighborhood and artificial neural network classification methods is assessed. Evaluation of data mining methods presents high accuracy and performance of these methods in assessment of water quality levels. In this study, the aforementioned methods are ranked by accuracy using kappa and error rate statistics. Careful examination of the results demonstrates that the support vector classifier which uses kernel functions is highly capable of solving various problems, and with an average ranking of 1.25 is the most efficient mining method followed by K- nearest neighborhood method with an average ranking of 1.75 and artificial neural network with an average rank of 2. These are also suitable methods for determining water quality classification.