Groundwater quality evaluation using a classification model: a case study of Jilin City, China

Using the MATLAB™ platform, groundwater quality in Jilin, China, is evaluated by employing integrated- and automation-type models. Genetic algorithm (GA), particle swarm optimisation, and support vector machine (SVM) theory are coupled in the model to form two-layer loop nesting. By using a GA, the surrounding loop enters the inner loop by choosing some factors from all measured evaluation factors. The inner loop is mainly composed of the SVM model. The inner loop feeds back the fitness function value of the GA obtained by weighting the model classification accuracy, and the reduced dimensions of each evaluation factor, to the surrounding loop. This aims to adjust the direction of evolution of the GA and eliminate evaluation factors with redundant, or sparse, information. The established model is applied to evaluate groundwater quality in Jilin and reduces 16 original evaluation factors to nine through a dimensionality reduction method. The training and verification sets constructed in the model exhibit more than 95% accuracy. Among the 183 wells used for monitoring groundwater in Jilin, the numbers of I-, II-, III-, IV-, and V-type monitoring wells are two, 96, 61, 20, and four, respectively. Compared with ordinary methods of evaluating water quality, the method integrates data selection and data processing instead of performing it in two successive substeps. The method exhibits the significant effect of dimensionality reduction on the number of its evaluation factors and also shows accurate evaluation results for water quality samples. Moreover, the method’s ability to be applied in many conditions provides a good basis for its use in various classification problems including water quality evaluation.