Using Machine Learning Algorithms for Identifying Mothers Who May Mutilate Their Daughters

Emmanuel Olamijuwon, University of St Andrews
Olaide Ojoniyi, University of the Western Cape
Jeremiah Olamijuwon, University of the Witwatersrand
Sidumo Masango, University of Eswatini
Clifford O. Odimegwu, University of the Witwatersrand

Despite efforts aimed at eradicating female genital mutilation, this practice remains endemic in Nigeria and many parts of the world. Although, several studies have identified the correlates of female genital mutilation, recent advances in computational and social science research have provided new ways of identifying mothers who may mutilate their daughters. We used data from the Nigeria demographic and health survey (2013) to train five machine-learning algorithms to predict if a mother could mutilate their daughter. Our models comprised of Support Vector Machine (SVM), Classification Trees (CART), Naïve Bayes (NB), Linear Discriminant Analysis (LDA), and k-Nearest Neighbors (KNN). We externally validated the models in the 2016 Nigeria multiple indicator cluster survey. Our findings during external validation, suggests that the linear discriminant analysis gives the best accuracy (86%) while the k-Nearest neighbor model had the lowest accuracy (76%). The implications of these findings for policy and scholarship are discussed.

See paper.

  Presented in Session 143. Computational Approach (Social Media, Big Data…) To Population Studies In sub - Saharan Africa