Utilizing Generative AI for Enhanced Model Explainability and Effective Decision Support in HR
Introduction
In the “Human Resources Predictive Analytics” (HRPA) section, we discussed a project that uses predictive analytics to uncover hidden patterns and predict employee attrition, often revealing insights that conventional HR methods miss. However, this project’s complexity introduces a significant challenge: the model’s opacity.
In a field as human-centered as HR, understanding why an employee decides to leave is crucial for making informed decisions and adopting effective retention strategies. To address this, we developed two essential tools to assist HR professionals within the HRPA framework.
Explainable AI Agent
The explainability of the model is achieved through Shapley values, a concept from cooperative game theory that allocates contributions to each factor in a prediction. In HRPA, Shapley values reveal how much each variable, such as salary, performance, or work-home distance, contributes to predicting employee churn.
This conversational agent provides real-time explanations of Shapley values. It can be deployed on platforms like Telegram, Google Chat, or other messaging apps. HR professionals can request the Shapley values for a specific employee (identified by name or ID) or a group of employees (e.g., a specific team), and the bot retrieves the model predictions and breaks them down into understandable components. This agent enables HR professionals to understand the business value of Shapley values.
What-if Analysis Tool
Understanding prediction factors is crucial, but HR professionals also need to experiment with different scenarios. The What-if Analysis Agent complements the Explainable AI Agent by allowing HR teams to simulate the impact of changes in employee-related variables on churn predictions (e.g., if an employee’s salary is increased by 5%, how does it affect their likelihood of leaving?).
For instance, if the model predicts John Smith is likely to leave, HR can ask the Shapley bot to explain why. If salary is a significant factor, they might consider increasing it. But how much should it be increased? The Explainable AI Agent can’t answer that, but the What-if Analysis Agent can.
Conclusions
The combination of the Explainable AI and What-if Analysis agents empowers HR professionals to navigate the complex landscape of employee churn with confidence and precision. Shapley values help identify the most influential factors contributing to churn predictions for individuals and teams. The What-If Analysis tool allows HR to experiment with different strategies before implementation, seeing how adjustments in compensation, benefits, or policies impact churn predictions without taking real-world risks.
In a fast-paced business environment, retaining top talent is crucial. These tools offer a dynamic approach to HR management, providing the necessary insights and flexibility to proactively address employee churn, ultimately contributing to a more stable and productive workforce.
