Machine Learning Methods to the Prediction of Iranian Household Health Expenditures

Background and aims: Household health expenditures play a crucial role in achieving goodhealth outcomes, but it is not practical to view it in isolation without taking into account otherfactors that contribute to overall health. The provision of health care coverage is subject to multiplefactors that can impact the generation of good health outcomes, as well as household healthexpenditures. The objective of this study is to examine the influence of household expendituresand demographic factors (including characteristics of the household and the household head) onhealth expenditure behavior.Method: A cross-sectional and retrospective study was conducted to identify predictor factorsthat affect the health expenditures of households. The data used in this study was obtained fromthe Iranian Statistic Center (SCI) for the period between ۲۰۱۶ and ۲۰۱۸, which included a standardquestionnaire on the income and expenditure of both urban and rural households. The goalwas to select an appropriate machine learning model to determine the importance of predictorvariables. Two auto-classifier and auto-numeric models were implemented on the data in theIBM-SPSS-Modeler, and two neural network models and a general linear model were proposed.In order to data analysis, machine learning models including multilayer perception (MLP) andgeneral linear model (GLM) was used. The dependent variable was health expenditure whichwas divided into three categories: (۱) without expenditures; (۲) more than ۱۲۰۰۰۰۰۰۰; (۳) lessthan ۱۲۰۰۰۰۰۰۰. The independent variables were divided into two categories; demographic andexpenditures. Demographic variables were (۱) marital status, (۲) number of people who haveincome, (۳) region, (۴) age, (۵) number of educated people, (۶) family Size, (۷) gender. Expendituresvariables involving housing, food, education, Entertainment and cultural activity, others,Transport and communication, Clothing.Results: Out of a total of ۱۱۴,۶۸۸ households, ۵۷,۸۶۰ were located in urban areas, which accountsfor ۵۰.۵% of the total households. The remaining ۵۶,۸۲۸ households were situated in ruralareas, comprising ۴۹.۵% of the total number of households. The study’s findings on the relationshipbetween demographic variables and nonlinear and linear models reveal that the number ofeducated people had the highest weight value, accounting for ۰.۲۷, followed by province with۰.۲۳, and age with ۰.۱۵. On the other hand, the effect size was more favorable for the followingfactors: (a) marital status; (b) number of people who have income; and (c) region. The study’sfindings on the relationship between household expenditures and linear and nonlinear modelsshow that the factors with the highest weight values are as follows: (a) housing with ۰.۱۹, (b)food with ۰.۱۷, and (c) education with ۰.۱۴. Additionally, the conventional factors for effect sizeinclude clothing, housing, and food. We computed the accuracy of the model for each individualyear, as well as for the aggregate of three years combined. For demographic features, the model’saccuracy was ۶۸.۱%, ۷۰.۸% and ۶۷.۶% for the years ۲۰۱۶, ۲۰۱۷, and ۲۰۱۸, respectively, whilethe total accuracy for the three-year period was ۵۶.۳%. Conversely, the accuracy of the model forexpenditures features was ۶۷.۳%, ۷۰.۴% and ۶۶.۶% for the years ۲۰۱۶, ۲۰۱۷, and ۲۰۱۸, respectively,while the total accuracy for the three-year period was ۴۵.۶%.Conclusion: It was concluded that number of educated people, province and age identified asmore effectively demographic factor, and housing, food and education as more effectively expenditures variable which influence on health household expenditures. Policy makers can utilizethis information as evidence to address the impact of socio-economic factors on household healthand to determine the relative importance of each factor. The results indicated that utilizing datamining and artificial intelligence can enhance the efficacy of smart governance.

کلیدواژه ها:

Household expenditures ، health expenditure ، social determinants of health ، data mining ، machine learning ، neural network