Abstract:
The aim of this study is developing a predictive model for divorce of couples by employing a
machine learning approach and ensemble learning techniques. Within the scope of this study,
prediction was carried out by using design science research method based on court case
documents and derived attributes based on Ethiopian divorce problems. In the employed method
data acquisition, data preprocessing, dataset splitting, model training and testing then ensemble
different models and prediction was done. The study used 3469 (67%) of divorced and 1731 (33%)
married couples dataset from two courts of Oromia regional state zones of Jimma and Wollega
zone. The collected dataset include ‘personal information form’. which have "number of document,
name of the accuser, name of the defendant, type of marriage, duration of marriage, number of
children, date and year of accusation, marital status, and location" and also ‘divorce predictors
scale’. Which is derived and prepared based on secondary data of published articles that consists
five (5) subscales zero to four. The employed algorithms in the study are Supportive Vector
Machine, Logistic regression, Naïve Bayes, K-Nearest Niebuhr, Decision Tree, and Random
Forest. With the accuracy of 91.9, 78.7, 81.7, 91.9, 92.5, and 92.5 percent’s respectively. But, after
ensemble all models the accuracy obtained was 92.3%. Therefore, ensemble different model is
best way to decrease variance of single model and to be neither under-fitted nor over-fitted. The
derived attributes, scales and results of this study can be used by the stakeholders that have direct
contact with families such as the Ministry of family issues and they can deal with the Cause,
consequence, and percentage of divorce of couples of Ethiopia in their screening activities to
decrease divorce
