Prediction of Malaysian women divorce using machine learning techniques / Nazim Aimran … [et al.]

This paper discusses the performance of three machine learning techniques namely Decision Tree, Logistic Regression and Artificial Neural Network for predicting divorce among Malaysian women. Secondary data were obtained from the Fifth Malaysia Population and Family Survey (MPFS-5) conducted by the National Population and Family Development Board (LPPKN). The total number of instances in the dataset was 7,644 ever married Malaysian women aged 15 to 59 years old. Divorce is currently a serious problem among the Malaysian community due to various reasons. In 2019, the divorce rate in Malaysia rose by 12% from the previous year. During the first three months of the Movement Control Order (MCO), i.e. from March 18 to June 18, 2020, the Syariah Court of Malaysia recorded 6,569 divorce cases. Worse, a total of 90,766 divorce cases were recorded from January to October 2020. Six predictive models were used for comparison, namely Decision Tree (C5.0 and CHAID), Logistic Regression (Forward Stepwise and Backward Stepwise), and Artificial Neural Network (Multi-Layer Perceptron and Radial Basis Function). Among the six predictive methods, the Decision Tree model (C5.0) was found to be the best model in classifying divorce among Malaysian women. The accuracy of the C5.0 model was 77.96% followed by the Artificial Neural Network (Multi-Layer Perceptron) and Logistic Regression (Forward Stepwise) model (74.68% and 67.89%, respectively). The order of important predictors in predicting divorce among Malaysian women is the wives’ employment status (0.1531) followed by the husbands’ employment status (0.1396), type of marriage (0.1327), race/ethnicity (0.1327), distant relationship (0.1212), the wives’ qualification level (0.1115), age group (0.1053) and religion (0.0998).