Single record prediction

You can make prediction for a single employee who is already hired or will be hired in the future. For this use the Prediction page. All the fields are required to be filled in, most of them already have predefined values. It is possible that some values are missed in the drop-down menus. In such case, chose 'other' option: that will mean 'none of the known before'. The model is trained on the data with parameters that took place in the past. Any new option of a parameter, like a new country or language is not a mistake, but will be recognized as unknown. Not necessarily it will even distort the prediction.

Predefined grouped parameters like number of FTEs do not imply 'other' options. Everything what we have should fall into one of them.

Date of the employment and Date of the last project assignment may have quite high impact on the prediction. Both can be also in the future, just make sure that employment date is not later than last project date.

Last project assignment is the key concept of the model. That simply means that started working with a new project (project is equal to client). It can be completely new hire or somebody who moved to another project. Attrition is predicted for the period since last project, not first day of employment.

Prediction results are displayed on the same page almost immediately. History of the individual predictions is not saved, each prediction is independent and does not impact following ones.

The model is not supposed to predict attrition for those who already left the company. However, it can be interesting to evaluate the probability that this person would leave. The same concept is applicable for those who stays with the company, but last project start date is long from the past.

Prediction in a batch

Prediction can be done in the more efficient way than just for individual records. For this use the Batch Prediction page.

There are completely the same parameters as for individual prediction, but they should be prepared manually in a .csv file. Correctness of the prepared batch file is the key to the precise prediction. If any of the options is filled-in differently that it is prescribed, such parameter will not be taken into account by the model as 'yet unknown', which can distort the results.

The results of the batch prediction are displayed on the page. Also, it can be downloaded as a table.

Description of the results columns: