FFTrees: An R package to create, visualise and use fast and frugal decision trees

Play FFTrees: An R package to create, visualise and use fast and frugal decision trees
Sign in to queue


useR!2017: FFTrees: An R package to create, visuali...

Keywords: decision trees, decision making, package, visualization
Webpages: https://cran.r-project.org/web/packages/FFTrees/, https://rpubs.com/username/project
Many complex real-world problems call for fast and accurate classification decisions. An emergency room physician faced with a patient complaining of chest pain needs to quickly decide if the patient is having a heart attack or not. A lost hiker, upon discovering a patch of mushrooms, needs to decide whether they are safe to eat or are poisonous. A stock portfolio adviser, upon seeing that, at 3:14 am, an influential figure tweeted about a 5 company he is heavily invested in, needs to decide whether to move his shares or sit tight. These decisions have important consequences and must be made under time-pressure with limited information. How can and should people make such decisions? One effective way is to use a fast and frugal decision tree (FFT). FFTs are simple heuristics that allow people to make fast, accurate decisions based on limited information (Gigerenzer and Goldstein 1996; Martignon, Katsikopoulos, and Woike 2008). In contrast to compensatory decision algorithms such as regression, or computationally intensive algorithms such as random forests, FFTs allow people to make fast decisions 'in the head' without requiring statistical training or a calculation device. Because they are so easy to implement, they are especially helpful in applied decision domains such as emergency rooms, where people need to be able to make decisions quickly and transparently (Gladwell 2007; Green and Mehr 1997)
While FFTs are easy to implement, actually constructing an effective FFT from data is less straightforward. While several FFT construction algorithms have been proposed 15 (Dhami and Ayton 2001; Martignon, Katsikopoulos, and Woike 2008; Martignon et al. 2003), none have been programmed and distributed in an easy-to-use and well-documented tool. The purpose of this paper is to fill this gap by introducing FFTrees (Phillips 2016), an R package (R Core Team 2016) that allows anyone to create, evaluate, and visualize FFTs from their own data. The package requires minimal coding, is documented by many examples, and provides quantitative performance measures and visual displays showing exactly how cases are classified at each level in the tree.
This presentation is structured in three sections: Section 1 provides a theoretical background on binary classification decision tasks and explains how FFTs solve them. Section 2 provides a 5-step tutorial on how to use the FFTrees package to construct and evaluate FFTs from data. Finally, Section 3 compares the prediction performance of FFTrees to alternative algorithms such as logistic regression and random forests. To preview our results, we find that trees created by FFTrees are both more efficient, and as accurate as the best of these algorithms across a wide variety of applied datasets. Moreover, they produce trees much simpler than that of standard decision tree algorithms such as rpart (Therneau, Atkinson, and Ripley 2015), while maintining similar prediction performance.
References Dhami, Mandeep K, and Peter Ayton. 2001. "Bailing and Jailing the Fast and Frugal Way." Journal of Behavioral Decision Making 14 (2). Wiley Online Library: 141–68.

Gigerenzer, Gerd, and Daniel G Goldstein. 1996. "Reasoning the Fast and Frugal Way: Models of Bounded Rationality." Psychological Review 103 (4). American Psychological Association: 650.

Gladwell, Malcolm. 2007. Blink: The Power of Thinking Without Thinking. Back Bay Books.

Green, Lee, and David R Mehr. 1997. "What Alters Physicians' Decisions to Admit to the Coronary Care Unit?" Journal of Family Practice 45 (3). [New York, Appleton-Century-Crofts]: 219–26.

Martignon, Laura, Konstantinos V Katsikopoulos, and Jan K Woike. 2008. "Categorization with Limited Resources: A Family of Simple Heuristics." Journal of Mathematical Psychology 52 (6). Elsevier: 352–61.

Martignon, Laura, Oliver Vitouch, Masanori Takezawa, and Malcolm R Forster. 2003. "Naive and yet Enlightened: From Natural Frequencies to Fast and Frugal Decision Trees." Thinking: Psychological Perspective on Reasoning, Judgment, and Decision Making, 189–211.

Phillips, Nathaniel. 2016. FFTrees: Generate, Visualise, and Compare Fast and Frugal Decision Trees.

R Core Team. 2016. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/.

Therneau, Terry, Beth Atkinson, and Brian Ripley. 2015. Rpart: Recursive Partitioning and Regression Trees. https://CRAN.R-project.org/package=rpart.





Right click to download this episode

The Discussion

  • User profile image

    Fast and frugal decision trees seem to me to be the same as decision lists introduced by Rivest (1987).

    There are well known cases where decision lists face difficulty; for example they cannot represent the XOR function, however they are trained.

  • User profile image

    Is there an rpubs post on this? The o up right now seems to be a dummy link: <https://rpubs.com/username/project>. Thanks!

  • User profile image
    Sarah Grogan

    Great talk. FFTrees is a great package for creating decision trees that can be interpreted easily by others (without a computational background). I love the option to plot the ROC curve and compare the fast & frugal trees with models from other packages, and also make cost-informed decisions by including a vector of costs in the model. I can also see a lot of value in manually creating a tree, to compare prediction accuracy of models based on intuition to models based on statistics.

Add Your 2 Cents