Malaysia household incomes classification prediction with k-means clustering and fuzzy inference system

• Main Author: Hamzah, Nur Atiqah
• Format: Thesis
• Language: English; English; English
• Published: 2018
• Subjects: QA Mathematics
• Online Access: http://eprints.uthm.edu.my/20/1/24p%20NUR%20ATIQAH%20BINTI%20HAMZAH.pdf; http://eprints.uthm.edu.my/20/2/NUR%20ATIQAH%20BINTI%20HAMZAH%20COPYRIGHT%20DECLARATION.pdf; http://eprints.uthm.edu.my/20/3/NUR%20ATIQAH%20BINTI%20HAMZAH%20WATERMARK.pdf; http://eprints.uthm.edu.my/20/

The economy level of the citizen has become a main concern for Malaysia as a developing country to improve the living status. On this point of view, the household income data would be a very useful information to measure the economic status of the population in Malaysia. This study aims to build a classification prediction of household incomes using fuzzy inference system (FIS) from the K-means clustering outputs. Thus, this study focuses on three main objectives which are (a) To apply K-means clustering on household incomes data, (b) To propose the prediction of household incomes classification using FIS, and (c) To analyze and validate the classification solution for household incomes and to compare with discriminant analysis. Initially, the number of groups in the household income data is determined by using K-means clustering. Accordingly, the outputs from K-means clustering are used to identify the membership functions, namely, triangle, trapezoidal and Gaussian membership functions. Furthermore, FIS models for each membership function are built for the household income class prediction based on clustering outputs. For verification, the root mean square error (RMSE) value for each FIS model is calculated and the percentage of data correctly classified using the FIS models built is compared with the discriminant analysis output. As a result, it is found that Mamdani FIS model with Gaussian membership function is the best model with the RMSE is 1.0396, while the percentage of data correctly classified is 64.9989%. In conclusion, the classification prediction of household incomes discussed in this thesis could identify the predicted class of household income in a tractable way and the efficiency of the technique used in this thesis for classification prediction of household income is highly recommended.