Alkire-Foster oriented ensemble fuzzy inference system for urban poverty classification

Abstract

Malaysia is a developing country which relies on the monetary approach to measure poverty. The approach is simple to measure but it is insensitive towards changes of the poor in multiple dimensions such as education, health and living standards especially in urban areas. Several current issues in classifying the urban poor include rigid dichotomy of the poor and non-poor, unable to capture changes that happens in various sub-groups of urban poor population and misclassified poverty indicators. This study developed a multidimensional poverty measurement framework which integrated i) Alkire-Foster approaches in quantification of multidimensional urban poor, ii) Adaptive Neural Fuzzy Inference Systems (ANFIS) to predict classification of urban poor and resolve the misclassification of urban poor and iii) ensemble ANFIS. 300 questionnaires were distributed to targeted households in Bandar Tasik Selatan, Kuala Lumpur. This study started with a comparison of datadriven Fuzzy Rule-Based System (FRBS) with the domain expert comprising FRBS classification. Next, the Alkire-Foster method was introduced which included parameter selection, dual cut off identification and aggregation of the poor. Then, the ANFIS prediction was carried out using various ANFIS combination models such as Genfis 1, Genfis 2 and Genfis 3 to predict the classification of urban poor. This study proceeded to improve the classification by proposing the ensemble ANFIS that included ensemble weighting and ensemble integration method. The performance of this proposed framework was evaluated using Root Mean Square Error (RMSE), Mean Square Error (MSE), and R-Squared. For validation purposes, this study was reviewed by officers at the Zakat Collection Centre, Kuala Lumpur as the domain experts. The findings showed that the Genfis 3 using Fuzzy C-Means clustering algorithm in ANFIS outperformed all the ANFIS models, by obtaining the least MSE and RMSE values and highest R-Squared. These results included the Health dimension which was excluded in the current poverty measurement. Overall, this study has managed to address the urban poor classification by providing multiple dimensions of the poor and produce robust prediction results.