Successful management isn't a question of machine learning...
In the last few weeks of Spring Term, the students aren’t the only ones looking forward to summer. As I was listening attentively to the group presentations one Tuesday morning, I couldn’t help but wonder whether I would be heading off to the tennis courts that afternoon. As I received a call from a friend during the mid-morning break offering to play, I took stock of the weather: sunny, hot, windy, and slightly humid. The last fourteen times I had considered playing, the conditions weren’t quite the same. All other things being equal, can you predict whether I was on the courts that afternoon?
Successful management isn’t a question of machine learning, but of people taking the right decisions in the context of their work. Exploring this context can help you understand the nature of the challenges you face — are you working in a deterministic environment of perfect information, or a stochastic environment with missing pieces? Can you assume that the data at hand contains the desired outcome supervised learning) or not (unsupervised learning)? With what kind of data do you have to work (qualitative, quantitative, discrete, continuous, nominal, ordinal… …)? How much time are you willing to spend to find the answer, and for how good an answer are you looking (the confidence interval)? Business analytics is learning to understand the nature of the problems to be solved, and then using the appropriate methodologies to transform the data into individual and collective action.
Decision trees can be used to visually represent supervised learning environments using categorical or continuous data. Decision trees are predictive models that use esitmates and probabilities to improve managerial decision-making. When used correctly (with or without relying on software…), decision trees allow managers to use observations about a problem (represented by the branches) to draw conclusions about the appropriate course of action to be taken (represented by the leaves). In the tennis problem above, a simple decision tree can be used effectively to predict my behavior that afternoon.
Decision trees are extremely useful in reducing the uncertainty of courses of action in supervised learning environments. Think of the problem as a game of “Twenty Questions”. I have a secret (in what conditions will I play tennis) — you can use the data to formulate questions that can be answered by yes or by no (did I play tennis when…?). Choose each closed-ended question to gain as much insight into my behavior as possible (when I played tennis). The ranking of the questions is referred to as a decision list, and the value of each answer is the information gain. In the case above, two questions are sufficient to predict my behavior, both today and in the future.
The data provided here was supplied originally as an exercise by Written, Frank and Hall.[i] This context of this weather problem is described using four variables (outlook, temperature, humidity and windy) with categorical attributes (hot, mild, cool; etc.). The decision list begins by identify the variable (outlook) that when known will produce the greatest number of correct answers. The next step is to identify the variable, given the values of outlook will provide the greatest information gain (humidity). With this second question (variable), we have our answer — yes, I was playing tennis that afternoon because if was sunny and humid.
Improving managerial decision making is the heart and soul of the Business Analytics Institute. In our Summer School in Bayonne, as well as in our Master Classes in Europe, we put data science to work for you and for your organization. The Institute focuses on five applications of of data science for managers: working in the digital age, data-driven decision making, machine learning, community management, and visual communications. Data-driven decision making can make difference in your future work and career. Improving your ability to take the right decisions is only a click away.
First published on Medium
Lee Schlenker is a Professor at ESC Pau, and a Principal in the Business Analytics Institute http://baieurope.com. His LinkedIn profile can be viewed at www.linkedin.com/in/leeschlenker. You can follow us on Twitter at https://twitter.com/DSign4Analytics
[i] Written, I., Frank E. and Hal, M. (2010), Data Mining: Practical Machine Learning Tools and Techniques, Third Edition, Elsevier