Prediction of the Probability of Divorce Using Machine Learning Algorithms

Abstract

For a happy and healthy life, a happy married life is very much essential. But nowadays, divorce cases are increasing rapidly day by day. According to a study, the divorce rate worldwide was 4.08% per 1000 married people in May 2022. For this reason, there is a need for effective prediction of divorce rate which helps the marriage counselor or therapist to understand how serious a case is. For this research purpose, a dataset was collected from UCI repository and contains some data based on the questions asked to the couple and the answers they gave. First, the dataset was cleaned using different ranking methods such as Information Gain, One R, Gain Ratio and ReliefF. Using these ranking methods, the most important fields that really affect the divorce are selected. Then, different classification algorithms such as Logistic Regression, Naïve Bayes, SGD, Decision Tree, Random Forest and Multilayer Perceptron were used and compared to find the accuracy. These algorithms are used first with all fields and then with 6 and 7 fields. These algorithms are used for 50:50, 66:34, 80:20 training/testing split and for 10-fold cross validation. When 7-fields are combined and checked with all algorithms for training-test and 10-fold cross validation, then 100% accuracy was found in Decision Tree, Random Forest and Multilayer Perceptron algorithms.