Creation of A Fuzzy Neural Networks to Assess Environmental Safety

Abstract

© 2020. All Rights Reserved. The article focuses on the problem of assessing the level of the fuzzy phenomena as environmental safety territorial entities. Application of neural networks allows to overcome the lack of available information on the input and to carry out the correct assessment of the environmental safety of territorial entities. The expediency of solving the problem on the basis of machine learning methods -fuzzy neural networks - is substantiated. A set of initial data used to construct neuro-fuzzy models is described. Total 20 input parameters and one output parameter are used in the dataset. The input set of parameters was reduced to 8, and the amount of data for training was 1733 records after performing the appropriate data pre-processing procedures related to the exclusion of insignificant input parameters and after the reduction of the input feature space based on the correlation analysis results, as well as the exclusion of outliers in the data. The Rapid Miner Studio analytical platform was used to prepare the initial data. A specially developed software package was used as a tool for analysing the prepared data and forming a fuzzy model for assessing environmental safety; the software implements the process of training fuzzy neural networks forming a model of a collective of fuzzy neural networks and a fuzzy knowledge production base for classification. As a result of training, a fuzzy model with a knowledge base containing 13122 fuzzy production rules was formed. The results of testing the knowledge base showed its adequacy and the achieved classification accuracy of 95.33%. The achieved accuracy exceeds the accuracy of other classification models based on the same input data. Thus, the constructed fuzzy model can be effectively used to identify environmental issues.