Developing A Predictive Model For Fertility Preference of Women of Reproductive Age Using Data Mining Techniques

No Thumbnail Available



Journal Title

Journal ISSN

Volume Title


Addis Ababa University


Background: Fertility is one of the major factors that determine the overall size, distribution and/or structure of a population. High fertility in developing countries(particularly, in the poorest of those countries) poses detrimental consequences like a high fraction of women experiencing pregnancies of order five and above and a greater likelihood of short inter-pregnancy intervals. These are threat to the health of mothers and their children. At a macro-level, high fertility also contributes to high population growth which in turn results in slow economic growth, environmental degradation and unemployment, among others. Assessing fertility preference helps identify the proportion of women who demand for children and those who intend to limit childbearing. This aids in developing and implementing appropriate intervention programs for the purpose of achieving reductions in fertility levels necessary to slow population growth. Objective: To explore the possibility of applying data mining techniques in developing a model that can predict fertility preferences of women of reproductive age from EDHS2011 women’s survey dataset collect by CSA. Methodology: For this study, a six-step hybrid knowledge discovery process model was adopted. Through the steps, a dataset containing 15 attributes and 16515 records of women was constructed for building models. Results: Three data mining classification algorithms, J48, Naïve Byes and neural Network (Multilayer Perceptron), were tested using 10-fold-cross-validation. The classifiers were implemented on the dataset with all and selected features. Several experiments were constructed and the accuracy achieved on selected feature subset was 75.92%, 77.34%, 78.03% for Naïve Bayes, Multilayer Perceptron and J48, respectively. Conclusion: In this study, feature selection generally improved prediction performance of the classifiers. J48 model with accuracy of 78.03% was found to be relatively better predictor of fertility preference of women. This research study did indicate that data mining can be applied to women’s dataset to identify determinants of fertility preference and classify women according to their childbearing preferences. Age, number of living children, education, child death experience, marital status, sex of child and region are found to be the most important factors that determine fertility preference ofwomen.



education, child death experience, marital status, sex of child