Big Pharma Employee Retention - Predictive Analytics (Part II)

Globex Pharma, focused on better understanding the reasons behind employee turnover, has requested a statistical analysis of the possible reasons for attrition in their ranks to reduce the costs to the organisation.

The study uses machine learning algorithms in Logistic Regression and Decision Trees in conjunction with the Globex Pharma Employee Survey data to produce two models that give us insight into the likely characteristics of an employee that has left Globex since the 2023 survey.

Introduction

Since the Globex Employee Survey in 2023 16.2% of Globex’s one thousand employees have left the organisation resulting in lost sales and increased costs in recruiting, induction and onboarding replacements with the total cost difficult to quantify.

Fig. 1 Employee Attrition Globex Employee Survey 2023

With a view to reduce the economic impact of attrition moving forward, we will complete a statistical analysis resulting in recommendations aimed at reducing attrition in the organisation, and in turn reduce the costs involved.

Objective

The goal is to answer the question:

“What attributes of an employee increase the likelihood of leaving Globex Pharma?"

Overall Methodology

The first model will follow the methodology of Logistic Regression, the second of Decision Trees.

Logistic Regression is a supervised machine learning algorithm utilised in predicting outcomes of a binary nature, such as yes/no or true/false, from a set of predictor variables. (“Logistic regression | Definition & Facts | Britannica,” 2024)

Decision Trees use a flowchart like structure to make decisions towards a prediction, like branches in a tree or a fork in the road, used to predict the likelihood of an outcome occurring such as an employee has left or not left. (“Decision Tree,” 2017)

For both methods, we have taken the twenty-three questions from the survey and using the popular R programming language to build a series of models based on different combinations of the survey question.

The survey questions are called “predictor variables” in that we will use them to predict the target variable “Attrition” – whether an employee will stay or leave.

Next step is, for each unique combination of predictor variables we will compare the accuracy of the model’s predictions.

Predictive Model One – Logistic Regression

“What attributes make an employee more likely to leave Globex Pharma?”

Predictive Model Two – Decision Tree

"What makes an employee more likely to leave Globex Pharma?”

Conclusion

Using the data the models have indicated there is a link between overtime performed by an employee, the number of years an employee has been in the workforce being under 3 years, marital status being single and being a frequent business traveller to name a few.

Further to those factors, there are high income earners particularly in the Sales departments that have no stock options which additionally correlated to higher levels of turnover in that group.

We have provided recommendations based on the models to reduce overtime, develop a strategy to retain younger employees, review all stock option allocations for consistency in the organisation, perform independent exit interviews moving forward and a review of the travel needs for employees, particularly in the sales department.

A focus on these areas should lead to improved attrition and reduction in the economic impacts for Globex Pharma. An additional review in 6-12 months should be undertaken to review the effects of the efforts laid out in this document for efficacy.

Review the full business report

Access R Code here