Forecasting Competing Risks for Navy Personnel Management WEAI 2021

June, 2021
IDA document: d-22651
FFRDC: Systems and Analyses Center
Type: Documents , Human Capital
Division: Strategy, Forces and Resources Division
Jay Dennis, Julie A. Lockwood, Rachel G. Augustine, Michael R. Guggisberg See more authors
Accurate, high-fidelity predictions of complex events can provide important information in many research, managerial, and operational contexts. Our open-source software package, the Finite-Interval Forecasting Engine (FIFE) provides a flexible, machine learning toolkit for forecasting panel attrition. Constructed to facilitate reuse and adaptation, FIFE is not bound to a single use case: it can accommodate any process wherein a subject—such as a person, group, or equipment item—is observed over multiple periods before potentially transitioning to one or more exit states. The algorithms within FIFE build on and generalize traditional tools for survival analysis in a machine learning context. FIFE can be used for both binary and competing risk survival analysis. It also includes a suite of built-in tools for data preprocessing, metrics, hyper parameter optimization, and visualizations.