A Real-time employee attrition prediction and risk scoring system

Abstract

Human Resource (HR) analytics is increasingly being explored around the globe for its potential in addressing employee attrition. Globally, the rate of attrition has been estimated to be about 25% higher in comparison to the pre-pandemic era. The effects of employee attrition including the loss of valuable talent and incurring costs for recruitment and on-boarding of new talent has been felt by companies in different sectors globally. Previous studies have made considerable efforts in not only understanding the concept of employee attrition but also in its early detection. This study aims to advance previous research by moving beyond merely identifying an effective machine learning technique to implementing the model that enables the human resource team to understand and assess employee attrition risk in real-time. This study provides a focus on three specific objectives that utilize human resource analytics approaches to understand the concept of attrition. Firstly, the study aims to use statistical approaches to analyze and identify the factors influencing employee attrition. Secondly, it aims to evaluate the effectiveness of machine learning algorithms in predicting employee attrition. Ultimately, the development of a system that predicts employee attrition and generates risk scores in real-time using relevant HR data marks a pivotal milestone for this study. Generalized Linear Model with interaction terms is the statistical approach which was utilized to assess the contributors of employee attrition. Job satisfaction, job involvement, years at company and monthly income were statistically significant thus are attributed to an employee’s decision to quit or stay. In this study, a performance evaluation and comparison of XGBoost, Random Forest (ensemble techniques) and Support Vector Machine, K- Nearest Neighbors as well as LogisticRegression machine learning models was conducted. Leveraging the employee records from the IBM dataset, Random Forest outperformed all the other models with an Accuracy of 80%, Precision of 91%, Recall of 85% and F1 Score at 88%. Insights from the first two research objectives were used to develop a real-time employee attrition and risk scoring tool. The solution provided under this study can be utilized in companies to provide data driven insights on attrition of their employee base. This study provides invaluable insights that can be used by various stakeholders including but not limited to, companies, data solution providers and the government to provide proactive measures to address attrition such as salary adjustment and management of employee work involvement. In conclusion, this study has contributed to the various on-going human resource analytic research which can be incorporated within organizational systems to address employee attrition and reduce costs incurred in recruitment and training of new talent.

Keywords: Attrition, Human resource, Machine learning, Risk Scoring.