Machine learning survival analysis on couple time-to-divorce data

Abstract

Marriage life does not always last harmoniously and occasionally can lead to divorce. The trend for the last three years since 2019 shows that divorce cases in Palangka Raya occur with a fluctuating trend that has recently been increasing. This research used a machine learning method called Survival Support Vector Machine on the divorce dataset in Palangka Raya. This research developed a feature selection technique using backward elimination to determine the factors influencing the couple’s decision to have their divorce registered in the religious court. The backward elimination method yielded the variables contributing to divorce: the number of children, the defendant's occupation, the plaintiff's age at marriage, the cause of divorce, and the defendant's education. Based on the comparison of the survival model performance between the Cox proportional hazard and the Survival Support Vector Machine, it was found that the latter was better since it had a higher concordance index and hazard ratio, which were 61.24 and 0.54, respectively. Thus, 61.24% of divorce cases were classified precisely by SUR-SVM in terms of the time sequence of events. Moreover, the hazard ratio of 0.54 indicated that the divorce rate of couples with censored status was 0.54 times than that of couples with failed/endpoint status.