FFTrees: A toolbox to create, visualize, and evaluate fast-and-frugal decision trees

Abstract

Fast-and-frugal trees (FFTs) are simple algorithms that facilitate efficient and accurate decisions based on limited information. But despite their successful use in many applied domains, there is no widely available toolbox that allows anyone to easily create, visualize, and evaluate FFTs. We fill this gap by introducing the R package FFTrees. In this paper, we explain how FFTs work, introduce a new class of algorithms called fan for constructing FFTs, and provide a tutorial for using the FFTrees package. We then conduct a simulation across ten real-world datasets to test how well FFTs created by FFTrees can predict data. Simulation results show that FFTs created by FFTrees can predict data as well as popular classification algorithms such as regression and random forests, while remaining simple enough for anyone to understand and use.

Combining white box models, black box machines and human interventions for interpretable decision strategies

“…In cases where both measures are deemed equally important, the sensitivity weight is simply set to 0.5; the so-called bacc. In this article, we use the FFTrees toolbox written in the R Language by Phillips et al (2017) and choose their ifan optimization algorithm as our benchmark. As explained by the authors, the ifan optimization algorithm assumes independence between cues and uses a brute-force method to optimize the decision thresholds and directions for each cue, ranking them from the most significant to the least significant (Phillips et al, 2017).…”

“…In this article, we use the FFTrees toolbox written in the R Language by Phillips et al (2017) and choose their ifan optimization algorithm as our benchmark. As explained by the authors, the ifan optimization algorithm assumes independence between cues and uses a brute-force method to optimize the decision thresholds and directions for each cue, ranking them from the most significant to the least significant (Phillips et al, 2017). After creating a set of several trees with different exit structures, these trees are then pruned to remove non-discriminant nodes; the tree with the highest accuracy measure is finally selected.…”

“…SelecƟon of the tree with the highest balanced accuracy F 4: ifan algorithm, adapted from Phillips et al (2017).…”

“…The dataset was divided into two subsets: a training set (80% of observations) for finding the optimal parameters, and an independent test set (20% of observations), displayed in the following figures. We apply the ifan algorithm of the standard Fast and Frugal Tree developed by Phillips et al (2017). Figure 5 displays the tree with the best "balanced accuracy" (bacc), i.e.…”

“…The overall accuracy statistics in the testing data are visible in the bottom panel from both the confusion matrix and the Receiver Operating Characteristic (ROC) curve. The ROC curve illustrates the trade-off between sensitivity (sens) and specificity (spec) of different algorithms, namely, 10 models coming from the FFT as well as 5 other competing models: the standard decision tree (CART), using the rpart package (Breiman, Friedman, Olshen & Stone, 2017); the logistic regression (LR), using the stats package (Phillips et al, 2017); the L2 regularized regression (RLR), using the glmnet package (J. Friedman, Hastie & Tibshirani, 2010); the random forest algorithm (RF), using the random Forest package (Breiman, 2001); and the Support Vector Machine (SVM) algorithm, using the default methodology in Karatzoglou et al (2006).…”