An adaptive neuro-fuzzy inference system (ANFIS) approach for measuring country sustainability performance

With the increasing demand for sustainable development, many international institutions and governments are seeking a balance between the environment, society and economy. With the aim of understanding and monitoring sustainability performance, various sustainability assessment methods have been developed. Fuzzy logic theory has been widely used for sustainability assessment. Good as these approaches are, there are criticisms that most studies use pre-defined simple linear membership functions (triangular or trapezoidal) and fuzzy rules, which are largely derived from experts' knowledge. However, sustainability is a very complex, multi-criteria issue, which contains various complex non-linear relationships. Moreover, it is time-consuming to find out the optimal membership functions and rules based on the expert knowledge. Therefore, it becomes necessary to explore a new approach for induction of membership functions and fuzzy rules. This paper introduces the adaptive neuro-fuzzy inference system (ANFIS) approach for country level sustainability assessment. The membership functions and fuzzy rules are generated from 128 training samples. The assessment results are close to the SAFE, Sustainability Assessment by Fuzzy Evaluation, model. Furthermore, three different types of non-linear membership functions, including Gaussian, bell-shaped and sigmoidal, are tested. The Gaussian membership function is the best one for country sustainability assessment. This study explores sustainability assessment, and results show that, by using appropriate training data, the ANFIS method is effective to measure the countries' sustainability performance. Using ANFIS, assessment accuracy can be further improved through appropriate selection of training samples using alternative data from UN-Habitat, or World Bank, or even new data sets.