EVALUATION OF THE PERFORMANCES OF ARTIFICIAL NEURAL NETWORK AND COX PROPORTIONAL HAZARD IN MODELING TIME TO FEMALE GENITAL MUTILATION IN NIGERIA

Abstract

Introduction: Female genital mutilation is a crime against womanhood, posing a great health and financial burden to individuals, families and the society. This study evaluated the performances of Artificial Neural Network (ANN) and Cox Proportional Hazard (CPH) in modeling time to Female Genital Mutilation (FGM) in Nigeria. Methods: Risk factors of FGM from birth to the age or period at which women were exposed to cutting (in years) were modeled using Kaplan Meier method (Non-parametric), Cox proportion hazard model (parametric), and neural network (parametric). Data for this study was extracted from the 2018 Nigeria Demographic Health Survey (NDHS) using multistage stratified random sampling of households of women of ages 15-49. The outcome of interest was whether a respondent has undergone FGM or not as of survey time. The risk factors that were considered in this study were education, religion, residence, ethnicity, daughter circumcision, region, religious belief, opinions about FGM and wealth index. Results: The findings from the ANN model suggested that type of residence, level of education, opinion of the women on FGM and ethnicity were the most important predictors of FGM. The prevalence of FGM among women with primary and tertiary (higher) qualification was 38.42 and 26.62 respectively. Yoruba ethnic group had the highest prevalence with 51.31. The prevalence rate of FGM among women in urban residence (35.61) was higher compared to rural residence. Southern region had the highest prevalence rate with south east and south west having 47.07 and 46.1 prevalence rate respectively Women from urban residence had 13.46% higher hazard of being circumcised compared to women from rural residence. Women with secondary education had 18.66% higher hazard of being circumcised compare to women with tertiary education while women with primary and no education had 73.23% and 50.72% lower hazard of being circumcised compared to women with tertiary education. The comparison between the CPH and ANN model suggested that CPH model was better in terms of classification ability than ANN because it has the higher AUC value of 0.1908 compared to 0.0989. Conclusion: Artificial Neural Network and Cox Proportional Hazard models are both appropriate for predicting time to FGM and they are recommended for predicting and determining the influence of risk factors on the time to female genital mutilation