Machine Learning in Human Resources
Algorithm: one of the best-sounding words in the English language and one of the most powerful problem-solving tools for decades.
With an increase in computing power over the past several decades, came an increase in the number of ways to use algorithms to solve problems previously too computationally intensive to solve. Today, we reference a subset of these algorithms as machine learning.
Machine learning (ML) is the process of teaching a computer system how to make accurate predictions when it is fed data. Essentially, the computer is being programmed how to learn based on a set of training data.
Examples of Machine Learning Being Used in Human Resources
To date, it is difficult to find a wide range of examples of machine learning being applied to HR. Below are a few examples for reference.
Job candidate identification: Glassdoor and LinkedIn use ML algorithms to look for candidates.
Google has tools to analyze job candidate attributes and proactively present them with positions aligned to their skills, experience, and personality.
JPMorgan began testing machine learning algorithms that identify rogue employees in 2015.
Chatbots are using ML to provide employees with answers to popular HR questions such as, “What are my medical benefits?” They can also be used to identify problems quickly. For example, a sudden influx of questions regarding late expense reimbursement can identify an issue with the reimbursement process.
Three Types of Machine Learning
Machine learning can be broken down into three categories depending on the goal of the algorithm.
Supervised Learning (We know what we want to predict)
An algorithm is used to predict a target variable (dependent variable) based on a given set of predictors (independent variables). A function is generated that maps inputs to desired outputs and the training process continues until the model achieves a desired level of accuracy on the training data.
Examples: Regression Analysis, Decision Trees, Random Forest, and K-Nearest Neighbour.
Unsupervised Learning (We don’t know what we want to predict)
This category of algorithms is used to cluster data into groups, for example, grouping customers for the purpose of customized marketing initiatives.
Examples: Apriori algorithm and K-Means.
Reinforcement Learning (Gaining knowledge for decision making)
The computer is exposed to an environment where it trains itself continually using trial and error. It learns from experience and tries to capture the best possible knowledge to make accurate business decisions.
Example: Markov Decision Process.
The Risks of Machine Learning
Machine learning is impacting an increasing number of decisions and comes with an increasing list of risks that must be addressed.
Poor Value Alignment
When topics become popular in the business world, companies want to appear “in the know.” This often results in unwise spending of departmental budgets. In the case of ML, it results in the initiation of ML projects using valuable resources without producing value from the solution. The same problem occurred when HR Analytics rose in popularity.
Under-estimation of ML Costs
The cost of developing a ML model is frequently underestimated. Even after the model is built and deployed, it still requires ongoing maintenance.
Unintended Consequences
I have previously warned that selecting metrics must be done with care in order to avoid unintended consequences. The same concept holds true for ML. ML models can exhibit unexpected behaviours.
Misinterpretation of the Model Results
Like statistics, ML algorithms are based on a set of assumptions. It is up to the user of these tools to understand whether the data they are analyzing meets the assumptions for valid usage.
Without knowing what’s behind any tool (mathematically), you can never be sure that the result it produced is valid. This can result in bad data-driven decisions.
In Summary
The opportunities for applying machine learning have increased substantially with increased computing power and a decrease in computing cost. However, applying ML algorithms to decision making must be done with attention to the detail behind the algorithm and with visibility into the risks associated with ML usage.
About Tracey Smith
Tracey Smith is the President of Numerical Insights LLC, a boutique analytics firm that helps businesses derive value from data and improve their bottom line. If you would like to learn more about how Numerical Insights LLC, please visit www.numericalinsights.com or contact Tracey Smith through LinkedIn. To read future posts, you can join Ms. Smith’s network by signing up here.