Deductive Data Mining

doi:10.1037/met0000252

Deductive Data Mining

2020-12-01

10

PRE

AI

DOI：10.1037/met0000252

Original

Original request for help

Save

Abstract

En 中文

Data mining methods offer a powerful tool for psychologists to capture complex relations such as interaction and nonlinear effects without prior specification. However, interpreting and integrating information from data mining models can be challenging. The current research proposes a strategy to identify nonlinear and interaction effects by using a deductive data mining approach that in essence consists of comparing increasingly complex data mining models. The proposed approach is applied to 3 empirical data sets with details on how to interpret each step and model comparison, along with simulations providing a proof of concept. Annotated example code is also provided. Ultimately, the proposed deductive data mining approach provides a novel perspective on exploring interactions and nonlinear effects with the goal of model explanation and confirmation. Limitations of the current approach and future directions are also considered. Translational Abstract Increasingly large data sets provide the opportunity to search and find complex associations between predictors and an outcome. Data mining methods can serve to investigate even extremely large numbers of predictors simultaneously, however, the results of these methods are not necessarily easy to interpret. This is due to the fact that most data mining procedures are designed to predict future data (e.g., predicting e-mail spam or credit scores) rather than to clarify which predictors interact or have linear or nonlinear effects on an outcome. We provide a general modeling framework, called deductive data mining, for researchers to organize and interpret the results of data mining algorithms. Deductive data mining is shown to be an effective modeling guide in order to extract and select predicts and their specific effects. Empirical examples, analyses, and Monte Carlo simulations are provided to illustrate deductive data mining as well as annotated R code.

Keywords:

deductive data mining

data mining

decision trees

model comparison

Deductive Data Mining

Abstract

Journal

Organization

Cited Papers