COMPARISON OF ARTIFICIAL NEURAL NETWORK AND LOGISTIC REGRESSION IN PREDICTING THE DESIRE TO STOP CHILDBEARING AMONG HIGH PARITY MARRIED WOMEN IN NIGERIA

Abstract

Nigeria is among the high fertility countries world-wide and the population growth rate is worrisome. Despite several programmes put in place by Government and international agencies to reduce fertility, many women still bear more than four children against the 1988 Nigeria population policy which emphasized that family should limit its size to four. In most analysis on fertility in Nigeria. Logistic regression and other models have been repeatedly in use and very few have used artificial neural network. This study therefore compared the performance of logistic regression (LR) and artificial neural network (ANN) in predicting the desire to stop childbearing among high parity women in Nigeria. This study utilized data(Couple's Recode) from 2013 Nigeria Demographic and Health Survey (2013 NDHS), a population-based cross-sectional study. Total children ever born was used as a proxy variable for parity, and a sample of 1919 women aged 15-49 was included in the analysis. Descriptive statistics were obtained for all variables. Chi-square test, multivariate logistic regression anti artificial neural network models were fitted, and compared using the confusion matrix, and the Receiver Operating Characteristic (ROC) curve. The mean age of the women was 34.4 ± 5.8 years, and the prevalence of the desire to stop childbearing among high parity women was 26.4%. Respondent's age, region, and number of surviving children were significantly associated with the desire to stop childbearing. AUROCs of LR and ANN models Were 0.86 (95% Cl: 0.84 - 0.89), and 0.54 (95% Cl: 0.51 - 0.57), respectively. The sensitivity and specificity of the LR model were 54.5% and 89.4%, respectively; while the sensitivity anti specificity of the ANN model were 16.6% and 78.8%, respectively. The desire to stop childbearing among high parity women is still a problem in Nigeria. In the model analysis. the LR model outperformed ANN model. Government should intensify efforts on programmes aimed at reducing fertility in Nigeria.