A NEW FRAMEWORK FOR GEOSPATIAL SITE SELECTION USING ARTIFICIAL NEURAL NETWORKS AS DECISION RULES: A CASE STUDY ON LANDFILL SITES
- 1School of Civil Engineering, Engineering Campus, Universiti Sains Malaysia, 14300 Nibong Tebal, P. Pinang, Malaysia
- 2School of Electrical & Electronic Engineering, Universiti Sains Malaysia, Nibong Tebal, 14300 Penang, Malaysia
Keywords: Geospatial site selection, Decision rule, GIS, Artificial neural network, Landfill, suitability map
Abstract. This paper briefly introduced the theory and framework of geospatial site selection (GSS) and discussed the application and framework of artificial neural networks (ANNs). The related literature on the use of ANNs as decision rules in GSS is scarce from 2000 till 2015. As this study found, ANNs are not only adaptable to dynamic changes but also capable of improving the objectivity of acquisition in GSS, reducing time consumption, and providing high validation. ANNs make for a powerful tool for solving geospatial decision-making problems by enabling geospatial decision makers to implement their constraints and imprecise concepts. This tool offers a way to represent and handle uncertainty. Specifically, ANNs are decision rules implemented to enhance conventional GSS frameworks. The main assumption in implementing ANNs in GSS is that the current characteristics of existing sites are indicative of the degree of suitability of new locations with similar characteristics. GSS requires several input criteria that embody specific requirements and the desired site characteristics, which could contribute to geospatial sites. In this study, the proposed framework consists of four stages for implementing ANNs in GSS. A multilayer feed-forward network with a backpropagation algorithm was used to train the networks from prior sites to assess, generalize, and evaluate the outputs on the basis of the inputs for the new sites. Two metrics, namely, confusion matrix and receiver operating characteristic tests, were utilized to achieve high accuracy and validation. Results proved that ANNs provide reasonable and efficient results as an accurate and inexpensive quantitative technique for GSS.