Tree Induction for Probability-Based Ranking

By Foster J. Provost, Pedro M. Domingos. 2003

607 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

Box drawings for learning with imbalanced data (2014, methodology): “…prune the decision trees beyond their built-in pruning as previous research shows that unpruned decision trees are more eﬀective in their predictions on the minority class (Provost and Domingos 2002; Chawla 2003), and because it would introduce more complexity that would be diﬃcult to control for.”
A systematic study of the class imbalance problem in convolutional neural networks (2017, methodology): “Since the basic version of ROC is only suitable for binary classiﬁcation, we use a multi-class modiﬁcation of it [67]”
EUSBoost: Enhancing ensembles for highly imbalanced data-sets by evolutionary undersampling (2013, background): “25 Minimum number of item-sets per leaf1⁄42 Confidence1⁄4Laplace Smoothing [42]”
Comparing naive Bayes, decision trees, and SVM with AUC and accuracy (2003, methodology): “(The difference between C4.4 and C4.5 is also significant, as observed by [ 18 ]).”
Adaptive Machine Learning for Credit Card Fraud Detection (2015, methodology): “The authors of [55, 103, 235] recommend to leave the tree unpruned and to use Laplace smoothing for obtaining posterior probabilities.”
Decision trees: a recent overview (2011, background): “Provost and Domingos (2003) showed that decision tree class probability estimates can be improved by skipping the pruning phase and smoothing the distributions by applying Laplace correction.”
C4.5 and Imbalanced Data sets: Investigating the eect of sampling method, probabilistic estimate, and decision tree structure (2003, methodology): “Thus, the frequency based probabilistic estimate can be written as: However, simply using the frequency of the correct counts (of classes) at a leaf might not give sound probabilistic estimates (Provost & Domingos, 2003; Zadrozny & Elkan, 2001).”
Predicting good probabilities with supervised learning (2005, background): “One way to deal with multiclass problems is to transform them to binary problems , calibrate the binary models, and recombine the predictions (Zadrozny & Elkan, 2002).”
Learning Decision Trees for Unbalanced Data (2008, methodology): “This was primarily motivated by the observations of Provost & Domingos [16].”
Forecasting cancellation rates for services booking revenue management using data mining (2010, methodology): “Provost and Domingos (2003) then developed the first probability estimation tree method: C4.”
Recent advances in decision trees: an updated survey (2022, methodology): “Since correctly estimating the confidence of a prediction is a crucial part of self-training, the authors employed techniques that have been previously reported to enhance the probability estimation capacity of DTs, such as Laplacian Correction (Provost and Domingos 2003), no-pruning (Provost and Do”
A Review On Evaluation Metrics For Data Classification Evaluations (2015, methodology): “In [13] [17] [31] the AUC was used to construct an optimized learning model and also for comparing learning algorithms [28] [29].”
Finding latent code errors via machine learning over program executions (2004, background): “(Decision trees can rank using boosting [7] or class probabilities at the leaves [ 16 ], but the decision tree implementation used in this paper does not rank points.) The decision tree technique is more likely to isolate clusters of like properties than SVMs, because each cluster can be separated b”
An experimental comparison of performance measures for classification (2009, background): “It has also been shown [11, 17] that although pruning usually improves accuracy in decision trees, it normally decreases the AUC of the decision trees.”
A survey on semi-supervised learning (2019, background): “This can be mainly attributed to the fact that most decision tree learning algorithms explicitly attempt to minimize the impurity in tree nodes, thereby encouraging small leaves and highly biased probability estimates (Provost and Domingos 2003).”