Deep learning-based labor relations prediction system with multi-source data fusion and early warning mechanisms.

Area of Science:

• Organizational Behavior
• Artificial Intelligence
• Data Science

Background:

• Workplace conflicts pose significant management challenges, often inadequately addressed by traditional methods.
• Labor relations are complex, multi-factorial, and require advanced analytical approaches.

Purpose of the Study:

• To develop and evaluate a deep learning-based early warning system for predicting workplace conflicts.
• To integrate heterogeneous organizational data for enhanced conflict prediction accuracy.

Main Methods:

• Developed an attention-based multi-modal fusion deep learning architecture.
• Integrated HR records, communication logs, performance data, surveys, and economic indicators.
• Combined MLP for cross-sectional data and LSTM for temporal patterns.

Main Results:

• Achieved 89.2% prediction accuracy, surpassing gradient boosting and tabular deep learning models.
• Demonstrated multi-modal integration improved performance by 4.5-12.8%.
• Real-world deployment showed 87.3% early warning success with 5-21 day lead times.

Conclusions:

• Deep learning with multi-modal data fusion offers a powerful tool for proactive workplace conflict prediction.
• The system provides valuable lead time for interventions, though false positives and performance during stress periods require further research.
• Methodological foundations for organizational risk assessment using multi-source data fusion are established.