Divorce Prediction Using Machine Learning Algorithms in Ha’il Region, KSA (Saudi Arabia)

Authors: Abdelkader Moumen, Ayesha Shafqat, Tariq Alraqad, Etaf Saleh Alshawarbeh, Hicham Saber & Ramsha Shafqat

Source: https://doi.org/10.1038/s41598-023-50839-1 (open access)

What Is Practically Relevant About This Study?

This study introduced the Divorce Predictors Scale (DPS), a tool I was previously unaware of. The authors tested the predictive power of various machine learning algorithms for later divorce, with one algorithm achieving particularly high accuracy. Why present this study? The idea of potentially conducting AI-assisted couple therapy in the coming years is intriguing. AI could listen and offer suggestions for anamnesis and later for appropriate interventions. Since reviewing this study, I’ve been contemplating whether and how the prediction of a high likelihood of divorce would help or possibly negatively impact my work.

Study Content: The researchers applied the “Divorce Predictors Scale” (DPS), based on the Gottman Couples Therapy model, in the Ha’il region, KSA (Saudi Arabia) and tested it with various machine learning algorithms. This study demonstrates how algorithms like Random Forest (RF) and Artificial Neural Networks (ANN) can identify relevant characteristics in a relationship that indicate an increased risk of divorce. By using such predictive models, couples at risk could be identified early in practice and interventions could be more targeted. The DPS could be used to detect specific conflict patterns, such as excessive criticism or lack of trust, and address these issues in therapy. The high prediction accuracy of up to 91.66% with the RF algorithm after feature selection is particularly impressive.

Methodology of the Study

The study was conducted in the Ha’il region, KSA, with a sample of 148 participants, of whom 116 were married and 32 were divorced. The researchers collected data using the “Divorce Predictors Scale” (DPS), which includes 54 different characteristics. This data was then analyzed using various machine learning algorithms, including Artificial Neural Networks (ANN), Naive Bayes (NB), and Random Forest (RF). Initially, a feature selection technique (Correlation Based Feature Selection, CBFS) was applied to identify the most important predictors of divorce. The selected features were then fed into the algorithms to test their prediction accuracy. The best results were achieved with the RF algorithm, which reached a prediction accuracy of 91.66%. The study also used Kappa values to assess the consistency of the predictions and ensure that the results were not merely coincidental. Overall, the methodology provides a robust foundation for quantifying divorce risks and developing targeted interventions.

Limitations

The study has several limitations. First, the sample size is relatively small and geographically limited to the Ha’il region, which may limit the generalizability of the results. Additionally, the study relies on self-reports from participants, which can lead to biases. Long-term studies with larger and more diverse samples would be necessary to further validate the results.

Peer-Review

This study was published in “Scientific Reports,” a peer-reviewed journal.

Disclosure

This text was generated with the help of ChatGPT and was editorially reviewed and edited. The study contents were not used as training material, and the analysis was conducted in compliance with current best practices regarding copyright.