Analysing the Possibility of Using SOM Models for Ranking Environmental Monitoring Indicators

Abstract

The article considers the features of multidimensional data clustering for environmental monitoring using the SOM neural network model, which allows identifying parameter groups with the maximum and minimum contribution, assessing the influence of each parameter on the formation of clusters based on the model weights. It is shown that SOM allows clustering multidimensional data in the process of environmental monitoring, and the Shapley algorithm determines the contribution of each feature taking into account all possible combinations. A technique for identifying the most significant parameters based on the weights of the neural network model is proposed. The neural network is trained and clustering is carried out, the influence of individual parameters on the formation of clusters is estimated, and groups of parameters with the greatest (and least) contribution to the model training process are identified. It is concluded that the significance of features when processing the original data by the PCA method and the proposed neural network model (SOM) is identical. The developed approach allows for the formation of well-founded recommendations on priority measures to improve the quality of water resources and identify critical pollutants.