a comparative study of performance of k-nearest neighbors and support vector machines for classification of groundwater

the aim of this work is to examine the feasibilities of the support vector machines (svms) and k-nearest neighbor (k-nn) classifier methods for the classification of an aquifer in the khuzestan province, iran. for this purpose, 17 groundwater quality variables including ec, tds, turbidity, ph, total hardness, ca, mg, total alkalinity, sulfate, nitrate, nitrite, fluoride, phosphate, fe, mn, cu, and cr(vi) from 41 wells and springs were used during an eight-year time period (2006 to 2013). the cluster analysis was used, leading to a dendrogram that differentiated two distinct groups. the factor analysis extracted eight factors accumulatively, accounting for 90.97% of the total variance. thus the variations in 17 variables could be covered by just eight factors. k-nn and svms were applied for the classification of the aquifer under study. the results of svms indicated that the best performed model was related to an exponent of degree one with an accuracy of 94% for the test data set, in which the sensitivity and specificity were 1.00 and 0.87, respectively. in addition, there was no significant difference among the results of different kernels, indicating that an acceptable result can be achieved by selecting the optimum parameters for a kernel. the results of k-nn showed roughly a lower efficiency compared with those of svms, where the sensitivity and specificity was reduced to 0.90 and 0.88, respectively, although the accuracy of the model was 93%. a sensitivity analysis was performed on the groundwater quality variables, suggesting that calcium next to nitrate were the most influential parameters in the classification of this aquifer.