Machine Learning – An Opportunity for New Directions and to Engage with New Areas

Emily Tucker

by Emily Tucker
Department of Industrial and Operations Engineering, University of Michigan

Machine learning (ML) is part-buzzword, part-powerful algorithmic toolbox.  It’s hyped to change the world, and people in many fields are beginning to use its techniques to understand their systems and make better decisions.  As operations researchers, we spend our careers developing models to provide insights.  There is a range of overlap between ML and operations research (OR), and I think the popularity of machine learning gives us the opportunity to work in new areas and solve broader, more complex problems.

One of my friends who is a biologist once told me about a conversation she had with her PhD advisor; while they were discussing possible methods, her advisor at one point responded with a wave of her hand, “Oh, why don’t you just do some machine learning on that.”  But neither knew much of the specifics! 

Even a passing mention can be an open door to a discussion of new techniques.  While folks may not have heard of operations research, we can discuss how “algorithms” can be used in other ways as people become more comfortable with the term. This increase in familiarity may help us work with our clients and collaborators to choose the best methods for their problem at hand, whether it involves clustering, or integer programming, or other techniques.

Many fields are beginning to use machine learning to solve complex problems. Astrophysicists are working to classify galaxies, and information scientists are analyzing literary text (“How Machine Vision Is Reinventing the Study of Galaxies,” 2015; Prospero, 2018).  The Seattle Seahawks, an American football team, use ML to try to prevent injuries (Soper, 2017).  Researchers are studying defensive strategies in the NBA (“Machine Learning Proves Useful for Analyzing NBA Ball Screen Defense,” 2016).  Recently, Nature published a paper that discussed how supervised machine learning and matching algorithms had improved the refugee assignment process (Bansak et al., 2018).

We don’t need to go far afield though to find new directions as machine learning naturally parallels many of the traditional applications of operations research.  Businesses have long struggled to manage “customer churn” by retaining and recruiting new customers, and ML algorithms can enhance traditional models by incorporating a wide range of data to understand customer dynamics, including analyzing click rates and detailed order histories.

 These insights can improve decision-making in perhaps unexpected ways (Neff, 2014).  Walmart has found that certain weather conditions correlate with food purchases; their steaks tend to sell when it is warm and windy whereas hamburgers do better when it is warm but less windy.  In turn, Walmart can tailor its advertising down to a zip code-level based on weather conditions, and as a result, sales have gone up.

 There have also been mountains of OR work in inventory management, and researchers are starting to use ML and deep learning techniques to improve order quantities and timing (Snyder, 2018).  Rather than separating the problems of estimating demand distributions and optimizing operational decisions, researchers have integrated the two by incorporating ML techniques and found that they can substantially reduce cost.

 ML is popular partly because of its impact and partly because it is relatively easy to implement, particularly in contrast to many optimization algorithms.  Within R or Python, if you download the appropriate package, only a few lines of code are needed to start using machine learning.

 A word of caution as we dive deeper, however ― the relative ease with which ML algorithms can be applied can obscure potential biases in their insights.  If we’re not careful, these biases can have major repercussions on perpetuating inequality (Mok, 2017) and may cause more problems than they fix!

 If you’re interested in learning more, several online resources ― including popular courses from Coursera (Ng, 2018) and helpful notes from Chris Albon (Albon, 2018) ― provide instruction in ML techniques.  Many INFORMS student chapters are also organizing sessions and workshops on machine learning. Check out the chapter highlights in this edition for a few examples.

 Machine learning and operations research are natural neighbors (one might even argue nearest-neighbors?), and I believe there’s an important and unique role for both.  Even basic familiarity with machine learning may open the door for OR professionals to start new and unexpected collaborations.


 Albon, C. (2018). Notes On Using Data Science & Artificial Intelligence To Fight For Something That Matters. Retrieved May 15, 2018, from
 Bansak, K., Ferwerda, J., Hainmueller, J., Dillon, A., Hangartner, D., Lawrence, D., & Weinstein, J. (2018). Improving refugee integration through data-driven algorithmic assignment. Science, 359(6373), 325–329.
 How Machine Vision Is Reinventing the Study of Galaxies. (2015). Retrieved May 15, 2018, from
 Machine Learning Proves Useful for Analyzing NBA Ball Screen Defense. (2016). Retrieved May 15, 2018, from
 Mok, K. (2017). Mathwashing: How Algorithms Can Hide Gender and Racial Biases. Retrieved May 15, 2018, from
 Neff, J. (2014). Cloudy With a Chance of Meatballs: How Weather Forecast Predicts Walmart’s Sales Outlook. Retrieved May 15, 2018, from
 Ng, A. (2018). Machine Learning. Retrieved May 15, 2018, from
 Prospero. (2018). Machines are getting better at literary analysis. Retrieved May 15, 2018, from
 Snyder, L. (2018). Machine Learning for Inventory Optimization. Retrieved May 15, 2018, from
 Soper, T. (2017). How the Seahawks use Microsoft’s new high-tech performance platform to prevent injury and plan practices. Retrieved May 15, 2018, from