machine learning (ml) refers to the study of methods or algorithms designed to learn the underlying patterns in the data and make predictions based on these patterns.1 ml tools were initially developed in the computer science literature and have recently made significant headway into business applications. a key characteristic of ml techniques is their ability to produce accurate out-of-sample predictions. academic research in marketing has traditionally focused on causal inference. the focus on causation stems from the need to make counterfactual predictions. for example, will increasing advertising expenditure increase demand? answering this question requires an unbiased estimate of advertising impact on demand. however, the need to make accurate predictions is also important to marketing practices. for example, which consumers to target, which product configuration a consumer is most likely to choose, which version of a banner advertisement will generate more clicks, and what the market shares and actions of competitors are likely to be. all of these are prediction problems. these problems do not require causation; rather, they require models with high out-of-sample predictive accuracy. ml tools can address these types of problems. ml methods differ from econometric methods both in their focus and the properties they provide. first, ml methods are focused on obtaining the best out-of-sample predictions, whereas causal econometric methods aim to derive the best unbiased estimators. therefore, tools that are optimized for causal inference often do not perform well when making out-ofsample predictions. as we will show below, the best unbiased estimator does not always provide the best out-of-sample prediction, and in some instances, a biased estimator performs better for out-of-sample data.2 second, ml tools are designed to work in situations in which we do not have an a priori theory about the process through which outcomes observed in the data were generated. this aspect of ml contrasts with econometric methods that are designed for testing a specific causal theory. third, unlike many empirical methods used in marketing, ml techniques can accommodate an extremely large number of variables and uncover which variables should be retained and which should be dropped. finally, scalability is a key consideration in ml methods, and techniques such as
[1]
Ron Kohavi,et al.
Wrappers for Feature Subset Selection
,
1997,
Artif. Intell..
[2]
M. Pontil,et al.
A Convex Optimization Approach to Modeling Consumer Heterogeneity in Conjoint Estimation
,
2007
.
[3]
T. Evgeniou,et al.
Disjunctions of Conjunctions, Cognitive Simplicity, and Consideration Sets
,
2010
.
[4]
Edoardo Amaldi,et al.
On the Approximability of Minimizing Nonzero Variables or Unsatisfied Relations in Linear Systems
,
1998,
Theor. Comput. Sci..
[5]
Galit Shmueli,et al.
To Explain or To Predict?
,
2010,
1101.0891.
[6]
Rich Caruana,et al.
An empirical comparison of supervised learning algorithms
,
2006,
ICML.
[7]
Lan Luo,et al.
Consumer Preference Elicitation of Complex Products Using Fuzzy Support Vector Machine Active Learning
,
2016,
Mark. Sci..
[8]
J. Friedman,et al.
Classification and Regression Trees (Wadsworth Statistics/Probability)
,
1984
.
[9]
Ronald L. Rivest,et al.
Constructing Optimal Binary Decision Trees is NP-Complete
,
1976,
Inf. Process. Lett..