Box drawings for learning with imbalanced data

By Siong Thye Goh, Cynthia Rudin. 2014

45 citations recorded by the source corpus.

The vast majority of real world classification problems are imbalanced, meaning there are far fewer data from the class of interest (the positive class) than from other classes. We propose two machine learning algorithms to handle highly imbalanced classification problems. The classifiers are disjunctions of conjunctions, and are created as unions of parallel axis rectangles around the positive examples, and thus have the benefit of being interpretable. The first algorithm uses mixed integer programming to optimize a weighted balance between positive and negative class accuracies. Regularization is introduced to improve generalization performance. The second method uses an approximation in order to assist with scalability. Specifically, it follows a \textit{characterize then discriminate} approach, where the positive class is characterized first by boxes, and then each box boundary becomes a separate discriminative classifier. This method has the computational advantages that it can be easily parallelized, and considers only the relevant regions of feature space.

Subfields: Artificial Intelligence, Evaluation & Benchmarks

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 Evaluation & Benchmarks

Citations in the corpus

Interpretable classifiers using rules and Bayesian analysis: Building a better stroke prediction model (2015, background): “Similarly to that work, Goh and Rudin ( 2014) fully learn sparse disjunctions of conjunctions using optimization methods.”
The Bayesian Case Model: A Generative Approach for Case-Based Reasoning and Prototype Classification (2014, neutral): “, [12, 32, 13, 23, 15]); and prototype- or case-based classifiers (e.”
Learning Certifiably Optimal Rule Lists (2017, methodology): “e continuous variables, we have found that it is not dicult in practice to construct a rule set that is sucient for creating a useful model. It may be possible to use techniques such as Fast Boxes (Goh and Rudin, 2014) to discover useful and interpretable rules for continuous data that can be used w”
Interpretable classification models for recidivism prediction (2015, background): “0 and CART, and has been observed by others (see, e.g., Goh and Rudin 2014).”
Learning customized and optimized lists of rules with mathematical programming (2018, methodology): “ABayesian counterpart to thework of Goh andRudin [26] is that ofWang et al [71], which proves some results analogous to those in Sect.”
Learning Optimized Risk Scores (2016, neutral): “[45] Goh, Siong Thye and Cynthia Rudin.”
Supersparse linear integer models for optimized medical scoring systems (2015, background): “Carrizosa et al. (2010), Van Belle et al. (2013) and Goh and Rudin (2014) take different but related approaches for constructing classifiers with binary threshold rules.”
Or's of And's for Interpretable Classification, with Application to Context-Aware Recommender Systems (2015, background): “The work of Goh and Rudin [18] is for real valued features only, whereas we focus mainly on categorical data, though binned or thresholded real-valued data would sufﬁce where the bins need not be exclusive.”
Data Augmentation in Emotion Classification Using Generative Adversarial Networks (2017, methodology): “…with imbalanced datasets, many methods were proposed, such as undersampling [17], synthesizing minorities [2], creating ‘box’ around minorities [8] and etc. Diﬀerent from these works, we aim to resolve this problem by generating data of minority classes from low-dimensional manifold, which…”
A Bayesian Framework for Learning Rule Sets for Interpretable Classification (2017, background): “Hauser et al. (2010); Goh and Rudin (2014) applied integer programming approaches to solving the full problems but would face a computational challenge for even moderately sized problems.”
Interactive and interpretable machine learning models for human machine collaboration (2015, unknown): “, [43, 134, 45, 90, 60]); and prototype- or case-based classifiers (e.”
Adaptive Hyper-box Matching for Interpretable Individualized Treatment Effect Estimation (2020, methodology): “Hyperboxes have been used extensively for regression (e.g., Peters 2011), classification (e.g., Xu and Papageorgiou 2009) and prediction (e.g., Goh and Rudin 2014) but notably, not for causal inference (Khuat et al. 2019).”