globusjournal

Family planning is one of the four pillars of safe motherhood initiative to reduce maternal death in developing countries. Despite progress in contraceptive use, unmet needs are wide open and fertility remains high. Ethiopia have a higher fertility rate which contributes to maternal and child health destitution. Therefore, the main objective of this study is to predict the likelihood of contraceptive method usage among women. The EDHS data, with 15, 683 records, used in this study was collected by central statistical agency. Data understanding and selection, data preprocessing such as handling of missing value, outlier detection, data discretization and feature selection are performed. Four experiments with two scenarios for each experiment were conducted and classification performances have been compared in order to determine optimal machines learning algorithms for predicting the likelihood of contraceptive method usage. All the models built from J48 Decision Tree classifier, Naïve Bayes classifier, support vector machine and Neural Network. Comparison based on Accuracy suggests that the support vector machine model performs slightly best in predicting the likelihood of contraceptive method usage with classification accuracy of 86.17%. Various variables were identified associated with contraceptive method usage like Knowledge of FP, Wealth index, Partner’s education level, number of living children, visited by health facility, place of residence, religion, Education level and hearing of FP messages through different media were significantly associated with contraceptive use. Based on the identified possible factors that determine the utilization of contraceptives and the identified target groups (women with low likelihood of utilizing contraceptive) program managers can design programs and help for proper implementation with regard to increasing utilization of family planning.

Keywords: Machine Learning, Decision Tree, Neural Network, Naïve Bayes Classifier, Support Vector Machine, Contraceptive Use.