MODELING TOTAL FERTILITY RATE IN INDONESIA: A COMPARISON OF FOURIER SERIES REGRESSION AND ELASTIC NET REGRESSION

Abstract

The Total Fertility Rate (TFR) describes population growth and socioeconomic development of a country. This statistic plays an important role in predicting future social and economic conditions. Indonesia has experienced a steady decline in TFR over the past few decades, which can be a serious problem if this trend continues. Therefore, the factor influencing the decline must be found. The independent variables include the percentage of women graduating high school, percentage of the poor population, poverty gap index, poverty severity index, prevalence of inadequate food consumption, proportion of people living below 50 percent of median income, unemployment rate, infant mortality rate, child mortality rate, and percentage of ever-married women aged 15–49 years using contraception methods. The aim of this study is to compare both Fourier Series Regression and Elastic Net Regression models to see which approximation can capture the TRF phenomenon that occurs in Indonesia and identify the causes of its decline. Fourier Regression is chosen because there is a repetition of patterns in several variables. Moreover, this data is experiencing multicollinearity; hence, Elastic-net Regression is the best way because this method overcomes the limitations of each Ridge and Lasso approach. These models are compared to see which is more suitable to capture the relationships between these factors and TFR. The best model obtained will provide a clearer understanding of Indonesia's underlying drivers of fertility decline. The result is that the Fourier Series Regression can model all variables better than the Elastic-net Regression, and the independent variables can explain the proportion of variance in the dependent variables by 97.91%, with all the independent variables significantly affecting the Total Fertility Rate.