Predicting Relationship Stability Using Communication Patterns

Abstract

Strong relationships are a key part of a happy and healthy life, both for individuals and society as a whole. This study looks at how the ways people communicate can reveal the likelihood of their relationships lasting over time. By focusing on things like how people talk and act during conversations, how they show emotions, and how they resolve disagreements, the research identifies patterns linked to stability. The findings emphasize that healthy communication—marked by respect, understanding, and effective problem-solving—is crucial for long-term success in relationships. These insights can help create better tools for counseling, education, and even technology to support stronger and more fulfilling connections. We also will be seeing which model works best on this dataset.

Introduction

This study explores how machine learning can predict relationship stability (e.g., between friends, partners, coworkers) by analyzing communication patterns—such as how often people talk, how long their interactions are, and the relationship type. Traditional research relied on surveys/interviews, but machine learning allows deeper insights from real communication data.

Objectives:

• Use machine learning to identify patterns that correlate with stable vs. unstable relationships.

• Evaluate multiple models to determine which best predicts relationship outcomes.

• Provide practical applications for counseling, workplace dynamics, and social media.

Methodology:

1. Data Collection & Preparation:

   • Source: Kaggle dataset.

   • Features include: message frequency, sentiment, response times, relationship type.

   • Data was labeled, encoded, and split into 80% training / 20% testing sets using stratified sampling.

2. Models Used:

   • Decision Tree: Interpretable model with depth limit of 5.

   • Random Forest: 100 trees, depth limit of 7; reduces overfitting.

   • Logistic Regression: Baseline linear model with max iterations set to 500.

3. Evaluation Metrics:

   • Accuracy, Precision, Recall, F1-score, Confusion Matrix

   • Special attention given to detecting unstable relationships (minority class).

4. Feature Importance:

   • Tree-based models and regression coefficients were used to determine the most influential features.

   • Visualizations (bar charts, decision trees) helped explain model reasoning.

Results:

• Unstable Relationship Detection: All models showed low recall and F1-scores for this class, likely due to class imbalance.

• Confusion Matrices: Revealed that models often misclassified unstable relationships as stable.

Key Insights:

• Stable relationships are easier to predict due to clear, consistent communication patterns.

• Unstable relationships are harder to detect, requiring better data balance and potentially more advanced techniques.

• Decision Tree model was most balanced and interpretable among those tested.

Challenges:

• Class imbalance (more stable than unstable relationships).

• Difficulty in capturing the nuances of unstable communication through existing features.

• Limited by reliance on labeled datasets that may oversimplify real-world complexities.

Future Directions:

• Apply balancing techniques (e.g., SMOTE) to improve model performance on minority classes.

• Explore advanced models like ensemble boosting or neural networks.

• Consider adding temporal features, emotional sentiment dynamics, or context for better prediction.

• Potential for integration into tools for therapy, workplace management, or social platforms.

Conclusion

This study explored the use of machine learning models— Decision Tree, Random Forest, and Logistic Regression—for predicting relationship stability based on communication pat- terns . The results indicate that all three models were effective in identifying stable relationships, with high accuracy and recall for the positive class. Among them, the Decision Tree Classifier performed the best, achieving the highest accuracy (0.89) and a balanced performance in both stable and unstable relationship predictions. However, the models faced challenges when predicting unstable relationships. Both Random Forest and Logistic Re- gression demonstrated low recall for unstable relationships, primarily classifying them as stable. This indicates a class imbalance issue, where stable relationships were more preva- lent in the dataset. While all models performed adequately for stable relationships, predicting unstable relationships with the same level of accuracy remains a challenge. Future work can focus on addressing this imbalance by employing techniques such as data augmentation or using algorithms specifically designed for imbalanced datasets. Ad- ditionally, incorporating more features or exploring other ma- chine learning techniques, such as ensemble methods or deep learning, could further improve the models’ ability to detect unstable relationships. In conclusion, while the models show promise in predicting relationship stability, further optimization and refinement are needed to enhance their accuracy in predicting unstable rela- tionships, which is crucial for the broader application of these models in real-world scenarios.

References

[1] ”Predicting Marital Stability: An Approach for More Characteristics,” 2024. [2] ”The Prediction of Marital Satisfaction Based on Com- munication Patterns, Attachment Styles, and Psychologi- cal Hardiness,” 2024. [3] ”Within-Couple Associations Between Communication and Relationship Satisfaction: A Daily Diary Study,” 2024. [4] ”Communication, the Heart of a Relationship: Examining Capitalization, Accommodation, and Self-Construal on Relationship Satisfaction,” 2024. [5] Nini2002, ”Relation Stability Dataset,” Kaggle, 2024. [6] ”Collaborative Communication Efficiency is Linked to Relationship Satisfaction and Stability,” 2024. [7] ”Emotion Dynamic Patterns Between Intimate Relation- ship Partners Predict Relationship Stability,” 2024. [8] ”Does Couples’ Communication Predict Marital Satisfac- tion, or Does Marital Satisfaction Predict Communica- tion?” 2024. [9] ”The Communication Patterns Satisfaction in Married Students,” 2024. [10] ”Positive, but Not Negative Emotions, Predict Intimacy in Couple Interactions,” 2024. [11] ”Implicit Theories of Relationships and Conflict Commu- nication Patterns in Romantic Relationships: A Dyadic Perspective,” 2024. [12] ”Marital Stability and Divorce Prediction Among Cou- ples: A Machine Learning Approach,” 2024. [13] ”Temporal Patterns Behind the Strength of Persistent Ties,” 2024, arXiv:1706.06188. [14] ”A Machine Learning Approach to Predicting Continuous Tie Strengths,” 2024, arXiv:2101.09417. [15] ”‘You Made Me Feel This Way’: Investigating Partners’ Influence in Predicting Emotions in Couples’ Conflict Interactions Using Speech Data,” 2024. [16] ”How Couple’s Relationship Lasts Over Time: The Role of Communication Patterns,” 2024.

Copyright

Copyright © 2025 Deepak Gahalot, Harshil Vasani, Prof . Harshil Kanakia. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.