How Voluntary Termination Is Predicted

People are often the enterprise's greatest asset, and their loss can be expensive for many reasons. System predictions can make you aware of potential issues and their likely causes so that you can address them.

For example, if an employee whose performance is predicted to be high is also identified as likely to leave voluntarily, you can consider changes to relevant factors, such as grade or location, to reduce the risk. Voluntary predictions appear on the Workforce Predictions work area.

Settings That Affect Prediction of Voluntary Termination

Predictions of voluntary termination are based on existing data from all work relationships. The process that collects relevant data and generates the predictions is Collect Data and Perform Data Mining for Predictive Analytics, which uses Oracle Data Mining and also predicts performance.

You can perform data collection either for the enterprise or for a specified manager assignment; however, the data-mining stage of the process is always performed on all of the latest available data.

The process has no default schedule. You are recommended to run the process:

• Weekly if the volume of relevant transactions in your enterprise (such as hires, terminations, and promotions) is high

• At least monthly if the volume of transactions isn't high

Schedule the process at a time of low system activity to avoid performance impacts.

How Voluntary Termination Is Predicted

Each prediction is a percentage value, which is the predicted probability of voluntary termination. It is calculated as follows.

1. For all employee work relationships, the process collects the values of a large set of attributes. The attributes include, for example, time in grade, current job, latest salary increase, performance rating, and number of sickness absences in the previous year. The attributes of interest include those most likely to show a correlation with voluntary termination. In some cases, simple values, such as manager name, are required; in others, such as percentage increase in sickness leave, the process calculates the values. Most of these attributes are held at the assignment level; therefore, for work relationships with multiple assignments, multiple values are collected.

   Contingent worker and nonworker work relationships are excluded.

2. The attribute values are passed to ODM, which identifies patterns and relationships in the data and builds a predictive model that captures the differences between employees who have terminated voluntarily and all other employees.

3. ODM makes predictions of voluntary termination for current employees according to the predictive model. For example, if voluntary termination is high in a particular job and department, current employees with that job in that department may have a greater risk of voluntary termination than workers in other jobs or departments.

   Each prediction relates to an employee assignment. For employees with multiple assignments, multiple voluntary-termination predictions are made (one for each assignment). If an employee reports to a single manager in multiple assignments, the manager sees multiple predictions for that employee.

These predictions enable you to identify employees at highest risk of voluntary termination. The absolute risk of voluntary termination for the high-risk group may still be low in percentage terms, but relative to that for other groups of employees, the risk is high.

Voluntary-termination predictions are available for both teams and individual assignments:

• Team predictions show the average risk for the team. They also show, for each factor, such as current salary or grade, the percentage of employee assignments for which the factor is the main risk factor.

• Individual predictions show the predicted risk for the employee assignment. The values of relevant factors, such as current salary, and the relative contribution that each factor makes to the prediction, also appear.