Can HR Analytics help you predict recruitment success?

The summary of a interpretable algorithm to predict the recruitment success

Organizations put their efforts mostly on employees’ success as their most important asset. Workforce planning and recruitment are among the most critical stages of an employee’s life-cycle, influencing the organizational return on investment directly.

HR analytics is an assistive tool alongside HR professionals to recruit the most suitable personnel for the vacant positions and avoid post-hire conflictions.

It will be surprising if we state that industry analysis, performance management, and employee retention are half critical as recruitment and workforce planning in the expected return of investment.

Hence we can presume that pre-hire approaches could be more influential than HR practices applied for retention.

The importance of interpretability in the algorithms

Different research groups have proposed several algorithms to predict an employee’s success. Still, the business does not consider HR analytics as an actionable tool. The reason could be the lack of prescriptive solutions. Mostly HR analytics algorithms focus on descriptive or predictive approaches. That’s why recently, a group of Industrial Engineers* developed a new methodology to prescribe the most suitable candidate for a vacant position in advance of their recruitment.

It is essential to keep in mind that data analytics is a decision support tool for recruiters to improve the efficiency of their decisions in real-world settings.

Prescriptive Approach in HR Analytics

The proposed algorithm* consists of a local prediction model for recruitment success per candidate and job title and a global optimization model. What makes this model unique is not the accuracy of the results but the interpretability of the outcomes, indicating the probability of successful recruitment per employee and job. Note that human decisions are inherently subjective, and sometimes bias may result in a lack of workplace diversity. So recruiters need an assistive tool that put aside the preferences and decide objectively. Prescriptive approaches would explain how the algorithm predicted performance and turnover before recruitment.

Objectivity in HR Analytics Results

The main goal is to improve hiring and enhance diversity in the workplace, so following the recruiter’s decision in the mathematical approach will not be efficient. Subjectivity and inaccurate judgements will affect the human’s ration besides the hiring process. To avoid subjectivity, developers added employee past or post-hire performance. Although post-hire reviews could give insight into retention activities, it may be too late to act upon recruitment errors. Hence early pre-hire foresight of employee success would reduce financial and social costs.

As mentioned, the algorithm would provide a decision support tool, increase diversity by 40%, and maintain a high recruitment success level.

Main Challenge in Machine Learning Applications:

Empirical Data

Usually, a data scientist would have access to a limited amount of data (features) from candidates before recruitment. The records of unrecruited or those who didn’t get the pick for an interview are not available. It is indispensable to acquire data from a wide range of applicants to extract significant HR analysis insight. Although to ensure the proposed classification method’s success, many features (more than 150) are adopted. Features involve numerical and categorical data, including age, gender, family and marital status, residence, nationality, education certificate, test results, background details, and potential assigned position. In addition to interviews and test scores, including leadership and language score, professional preferences, and more information about the job are the algorithm’s features.

Classification Method

As mentioned, HR analytics approaches have three main categories: descriptive, predictive, and prescriptive. Analysts mostly applied descriptive statistics, like hypothesis testing(t-test), analysis of variance, regression, and correlation analysis to describe HR data. HR analytics mainly use Machine Learning algorithms to find patterns and predict employee performance, turnover in the first year of employment, or recruitment. Decision tree and support vector machine(SVM) were amongst the most accurate classification method clarifying the root causes of turn over. Yet, they did not seem to be applicable in real situations. That’s why data scientist focused on interpretability instead of accuracy.

The Variable Order Bayesian Network (VOBN) is the required prescriptive methodology in an actual organizational environment that represents comprehensive results and provides HR professionals with a handful of prescriptions instead of predicting their actions. VOBN identifies context-based patterns supporting the organization in the recruitment process. It also extracts rules and measures for the recruiters without any background. Moreover, it calculates scores and specific insight into factors and root causes that affect recruitments’ success. The algorithm classified the results in successful and unsuccessful recruitment based on HR department recordings. If the employee left for natural reasons like quitting the job after a sufficient period, the algorithm would label the result as successful recruitment.

On the other hand, if the employee leaves the job after a short amount of time or he/she shows poor performance, the label would be unsuccessful. The algorithm labels the position changes based on the leading reasons. For instance, misfit is unsuccessful recruitment (negative), and promotion is positive.

Global Recruitment Optimization

The global optimization perspective considers multiple goals for various organizational stakeholders. The proposed classifying algorithm should satisfy the following requirements: demand, accuracy, and diversity.

Demand means the minimized difference between the required workforce and the actual number of recruited employees. Maximization of the sum of the successful recruitment’s probability in an organization is accuracy. And maintaining a heterogenous work environment or, in other words balancing diverse groups of employees is diversity. Global recruitment task is an optimization problem that needs a mathematical programming formulation.

HR analytics as a Supportive Tool for HR professionals

The early identification of a possible misfit can save a great deal of financial and social costs. What’s more, such HR analysis would provide actionable recommendations for preventive actions. HR professionals do not need deeper technical or machine learning knowledge to implement the suggested methodologies and make objective decisions.

*If you want to read more about the mentioned algorithm check the full paper here: Employees recruitment: A prescriptive analytics approach via machine learning and mathematical programming