Application of Machine Learning Techniques for Classifying Household Standards of Living in Sinnar State, Sudan: A Comparative Study of Discriminant Analysis and Decision Trees

Abstract

This study aims to classify households in Sinnar State, Sudan, into high, medium, and low standard-of-living groups based on selected economic, demographic, and social variables. Using a two-stage cluster sampling method, data were collected from 800 households across 23 administrative units. A structured questionnaire was used to gather primary data, and SPSS software was employed for statistical analysis. Discriminant analysis and decision tree models were applied to identify key factors affecting household classification and assess model accuracy. Results showed significant differences between groups, confirming the suitability of discriminant analysis. The most influential variables in the first discriminant function were income sufficiency, car ownership, and health expenditure sources. The decision tree model slightly outperformed discriminant analysis, achieving 72% classification accuracy compared to 71.6% from the discriminant model. Despite the close performance, both methods effectively distinguished household living standards.

The study concludes that advanced statistical techniques such as discriminant analysis and decision trees are useful for socio-economic classification and can support better-targeted development policies. It recommends applying these models to guide government interventions and focusing on income-generating initiatives to enhance household welfare.

Keywords