Information-theoretic feature selection for a neural behavioral model

Employers of hourly workers typically experience high employee turnover. Due to costs associated with: training, hiring and termination, the overhead from this high turnover rate is substantial. It is therefore desirable to construct employee selection procedures and analytic models to estimate the likely tenure of applicants for employment prior to a hiring decision. A critical component in the success of this effort to create a neural network model to estimate tenure was the application of information-theoretic feature selection. The benefits of this technique are demonstrated by comparison with results obtained using no feature selection and alternate methods of feature selection.