Last Updated on
Objective
HR analytics has been gaining its much deserved traction, enabling organizations to make data driven decisions (more than a passing fad, data driven decision making is also one of the key ingredients of building a high performance organization), and we, at The Brew, were approached by the client organization to help them de-clutter their attrition – to analyze attrition risk and factors which influences the turnover in an organization (the turnover costs at the organization had a definite area of improvement – also see The Brew’s Turnover Costs Estimator) using multiple data mining techniques (See all of works & applications of HR Analytics here).
In this case, different predictive models are used to understand the turnover phenomenon. The result is the analysis provides the recommendations which enhances the efficiency and effectiveness of human resource planning processes that are used to focus on the employee turnover problem.
This study is conducted on real data provided by a global IT organization, which is an Amsterdam based company. For presenting this report, the data received from client was fully anonymized to fulfill the privacy rules. The necessary information which is used to answer our research question was present in the data.
Attrition Risk Findings
The attrition risk factors which have highest influence on employee turnover are Age, Location, Currency, Base salary, Business level, FTE, In-position and In-service. The factors which have less impact in predicting the employee turnover are Hire type, Gender, Contract type and Functional area.
Demographic factors like age and location are strong predictors of employee turnover because the younger employees from age 18-25 are more likely to turnover than older employees. Since younger employees leave in early stages so in-position and in-service also have some effect on turnover. These results are consistent with a study on turnover rates conducted by Hill and Associates which found that young undergraduates, graduates and post graduates in the outsourcing business had changed their jobs at least once in the past three years (Link). The location in our analysis has an impact because most of the employees who leave the company are from UK.
The results of our predicting models indicate that the random forest works best on these datasets. It gives the highest accuracy and sensitivity values which means that it can predict the employee turnover and minority class (Left) more precisely. The highest achieved overall accuracy is (94.07%) on the Europe dataset and the highest sensitivity value is (81.17%) which means this model can predict the minority class with 81% accuracy and it will make only 3% (1-specificity) mistakes to predict the majority class as minority class. On the USA and Total datasets, the sensitivity value is low because fewer number of variables were available for these datasets.
The results also reveal that the Currency and Base Salary are among the most important factors because these factors were not available in the USA and Total datasets, so the accuracy of our model was quite low on these datasets. The location factor has also some effect because it was not available in the USA dataset and the accuracy of model was the lowest for this dataset.
Since the currency is one of the most important predictors of employee turnover, it can be explained by the fact that people who are working in UK and not getting salary in GBP are more likely to leave the company. Moreover, the people who are working as a part-time employee are more likely to leave the company as compared to a full-time employee. The salary of the part-time employee is also lower than the full-time employee.
Conclusions to mitigate attrition riisks
This study investigated the attrition risk factors of influence on employee turnover using the data mining techniques. Three important conclusions can be made from this research study. First, it finds the importance of prediction models which can be used to predict the employee turnover. By considering different models, it is investigated that the best prediction is possible using random forest.
Secondly, the identification of the important factors like age, location, currency, business level, in-position, in service and FTE are significant from a research perspective. Lastly, as this analysis is specific to the given dataset so these predictive accuracies can be used by client organization, providing this dataset and they can identify those employees who have turnover intentions even before they had made their final decision to leave.