Spatial Data Mining Model For Landfill Sites Suitability Mapping Based On Neural Networks And Multivariate Analysis

• Main Author: Abujayyab, Sohaib K. M.
• Format: Thesis
• Language: English
• Published: 2017
• Subjects: T Technology; TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.usm.my/45756/1/Spatial%20Data%20Mining%20Model%20For%20Landfill%20Sites%20Suitability%20Mapping%20Based%20On%20Neural%20Networks%20And%20Multivariate%20Analysis.pdf; http://eprints.usm.my/45756/
• Institution: Universiti Sains Malaysia

It is very crucial to have a precise suitability mapping workflow for new landfill sites in the development planning of municipal solid waste management systems. An appropriate siting of landfill sites will protect both environment and public health. However, the complexity in the process of suitability mapping that arises from the attempt to integrate information or decisions from different disciplines has affected the results and leads to inefficient landfill siting model. There are several Spatial Data Mining (SDM) methods and Multi Criteria Decision Analysis (MCDA) workflows that are currently available, but their application in landfill sites selection is limited and reveals a number of drawbacks. In this study, the enhancement of the SDM model was constructed to serve four purposes; (1) new workflow in creating suitability maps at the regional scale for solid waste planning based on neural network (NN); 2) a hybrid network that combines layer-recurrent network and cascade forward neural network to achieve high performance without requiring prior human knowledge; 3) a methodology for selecting the relevant input criteria for landfill GIS model based on multivariate analysis (MVA) methods for maximal performance; and 4) automating an ArcGIS neural network spatial data mining toolbox for mapping the suitability of landfill sites at a regional scale. A case study on landfill site selection in four northern states of Malaysia was conducted to demonstrate the validity of the new SDM model. A total of 31 criteria were pre-processed to establish the input dataset for NN modeling. From these, 22 criteria were adopted as input datasets after pre-checking for multicollinearity. The learned network was used to acquire the weights of the criteria. The optimum structure of the proposed network was selected using 600,000 use cases. Six MVA methods were employed to select the relevant criteria. Hybrid neural network was utilized as an evaluation method to select the optimal selection method and optimal training algorithm. The employment of automated toolbox is a straightforward process constructed from eight sub-tools to prepare, train, and processes the data. An accuracy of 99.2% was achieved for the test dataset. The final structure of the trained network was used to produce the suitability index map. The result showed that the LM training function with ‘Consistency-Subset-Eval’ selection method has efficiently identified 14 criteria with a performance accuracy of 99.2%. In addition, five out of the six methods has selected seven identical criteria that were most relevant. The workflow was found to be capable of reducing human interference to generate highly reliable maps. The developed network and the proposed workflow reveal the robust and the applicability of NN in generating landfill suitability maps and the feasibility of integrating them with existing MCDA workflows. The research outcomes show that the methodology of selecting and ranking criteria is quicker, economical, and precise. It can be an alternative to the existing time-consuming methodologies for selecting relevant criteria. Lastly, the automated model generated can certainly and effectively provides platform for decision makers to implement the developed workflow and methodology as well as the network. In conclusion, developed SDM model is recommended for long-term planning of solid waste management and to produce suitability maps for new landfill sites.