Predicting Company Quitting From Online Social Enterprise Activity

Modeling and predicting attrition in organizations has real-world business significance. In this paper, we take a novel approach of analyzing a corporate social network (Yammer) to predict if people are likely to quit their company. Via a data-driven approach, we compute a rich set of features derived from graph structure, content, and work practice characteristics derived from Yammer. Our experiment shows that the proposed data-driven approach can be used to predict employee quitting with a fair accuracy of approximately 68% and a moderately high recall rate of 62%. Given the difficulty of the quitting prediction problem, these accuracy and recall rates are fairly encouraging.