Daniel J. Lizotte

Jacqueline K. Kueper, Jennifer Rayner, Daniel J. Lizotte

Electronic health records (EHRs) include simple features like patient age together with more complex data like care history that are informative but not easily represented as individual features. To better harness such data, we developed an interpretable hybrid feature- and similarity-based model for supervised learning that combines feature and kernel learning for prediction and for investigation of causal relationships. We fit our hybrid models by convex optimization with a sparsity-inducing penalty on the kernel. Depending on the desired model interpretation, the feature and kernel coefficients can be learned sequentially or simultaneously. The hybrid models showed comparable or better predictive performance than solely feature- or similarity-based approaches in a simulation study and in a case study to predict two-year risk of loneliness or social isolation with EHR data from a complex primary health care population. Using the case study we also present new kernels for high-dimensional indicator-coded EHR data that are based on deviations from population-level expectations, and we identify considerations for causal interpretations.

Nathan Phelps, Stephanie Marrocco, Stephanie Cornell et al.

Reinforcement learning (RL) has helped improve decision-making in several applications. However, applying traditional RL is challenging in some applications, such as rehabilitation of people with a spinal cord injury (SCI). Among other factors, using RL in this domain is difficult because there are many possible treatments (i.e., large action space) and few patients (i.e., limited training data). Treatments for SCIs have natural groupings, so we propose two approaches to grouping treatments so that an RL agent can learn effectively from limited data. One relies on domain knowledge of SCI rehabilitation and the other learns similarities among treatments using an embedding technique. We then use Fitted Q Iteration to train an agent that learns optimal treatments. Through a simulation study designed to reflect the properties of SCI rehabilitation, we find that both methods can help improve the treatment decisions of physiotherapists, but the approach based on domain knowledge offers better performance. Our findings provide a "proof of concept" that RL can be used to help improve the treatment of those with an SCI and indicates that continued efforts to gather data and apply RL to this domain are worthwhile.

Nathan Phelps, Daniel J. Lizotte, Douglas G. Woolford

When modelling data where the response is dichotomous and highly imbalanced, response-based sampling where a subset of the majority class is retained (i.e., undersampling) is often used to create more balanced training datasets prior to modelling. However, the models fit to this undersampled data, which we refer to as base models, generate predictions that are severely biased. There are several calibration methods that can be used to combat this bias, one of which is Platt's scaling. Here, a logistic regression model is used to model the relationship between the base model's original predictions and the response. Despite its popularity for calibrating models after undersampling, Platt's scaling was not designed for this purpose. Our work presents what we believe is the first detailed study focused on the validity of using Platt's scaling to calibrate models after undersampling. We show analytically, as well as via a simulation study and a case study, that Platt's scaling should not be used for calibration after undersampling without critical thought. If Platt's scaling would have been able to successfully calibrate the base model had it been trained on the entire dataset (i.e., without undersampling), then Platt's scaling might be appropriate for calibration after undersampling. If this is not the case, we recommend a modified version of Platt's scaling that fits a logistic generalized additive model to the logit of the base model's predictions, as it is both theoretically motivated and performed well across the settings considered in our study.

Nathan Phelps, Daniel J. Lizotte, Douglas G. Woolford

There is a widespread and longstanding belief that machine learning models are biased towards the majority (or negative) class when learning from imbalanced data, leading them to neglect or ignore the minority (or positive) class. In this study, we show that this belief is not necessarily correct for decision trees, and that their bias can actually be in the opposite direction. Motivated by a recent simulation study that suggested that decision trees can be biased towards the minority class, our paper aims to reconcile the conflict between that study and decades of other works. First, we critically evaluate past literature on this problem, finding that failing to consider the data generating process has led to incorrect conclusions about the bias in decision trees. We then prove that, under specific conditions related to the predictors, decision trees fit to purity and trained on a dataset with only one positive case are biased towards the minority class. Finally, we demonstrate that splits in a decision tree are also biased when there is more than one positive case. Our findings have implications on the use of popular tree-based models, such as random forests.

Nathan Phelps, Daniel J. Lizotte, Douglas G. Woolford

When using machine learning for imbalanced binary classification problems, it is common to subsample the majority class to create a (more) balanced training dataset. This biases the model's predictions because the model learns from data whose data generating process differs from new data. One way of accounting for this bias is analytically mapping the resulting predictions to new values based on the sampling rate for the majority class. We show that calibrating a random forest this way has negative consequences, including prevalence estimates that depend on both the number of predictors considered at each split in the random forest and the sampling rate used. We explain the former using known properties of random forests and analytical calibration. Through investigating the latter issue, we made a surprising discovery - contrary to the widespread belief that decision trees are biased towards the majority class, they actually can be biased towards the minority class.

Brent D. Davis, Ethan Jackson, Daniel J. Lizotte

We present an algorithm, Decision-Directed Data Decomposition (D4), which decomposes a dataset into two components. The first contains most of the useful information for a specified supervised learning task. The second orthogonal component contains little information about the task but retains associations and information that were not targeted. The algorithm is simple and scalable. We illustrate its application in image and text processing domains. Our results show that 1) post-hoc application of D4 to an image representation space can remove information about specified concepts without impacting other concepts, 2) D4 is able to improve predictive generalization in certain settings, and 3) applying D4 to word embedding representations produces state-of-the-art results in debiasing.

Daniel J. Lizotte, Arezoo Tahmasebi

We develop and evaluate tolerance interval methods for dynamic treatment regimes (DTRs) that can provide more detailed prognostic information to patients who will follow an estimated optimal regime. Although the problem of constructing confidence intervals for DTRs has been extensively studied, prediction and tolerance intervals have received little attention. We begin by reviewing in detail different interval estimation and prediction methods and then adapting them to the DTR setting. We illustrate some of the challenges associated with tolerance interval estimation stemming from the fact that we do not typically have data that were generated from the estimated optimal regime. We give an extensive empirical evaluation of the methods and discussed several practical aspects of method choice, and we present an example application using data from a clinical trial. Finally, we discuss future directions within this important emerging area of DTR research.

Rhiannon V. Rose, Daniel J. Lizotte

When faced with a supervised learning problem, we hope to have rich enough data to build a model that predicts future instances well. However, in practice, problems can exhibit predictive heterogeneity: most instances might be relatively easy to predict, while others might be predictive outliers for which a model trained on the entire dataset does not perform well. Identifying these can help focus future data collection. We present gLOP, the global and Local Penalty, a framework for capturing predictive heterogeneity and identifying predictive outliers. gLOP is based on penalized regression for multitask learning, which improves learning by leveraging training signal information from related tasks. We give two optimization algorithms for gLOP, one space-efficient, and another giving the full regularization path. We also characterize uniqueness in terms of the data and tuning parameters, and present empirical results on synthetic data and on two health research problems.

Tameem Adel, Benn Smith, Ruth Urner et al.

We present a comprehensive study of the use of generative modeling approaches for Multiple-Instance Learning (MIL) problems. In MIL a learner receives training instances grouped together into bags with labels for the bags only (which might not be correct for the comprised instances). Our work was motivated by the task of facilitating the diagnosis of neuromuscular disorders using sets of motor unit potential trains (MUPTs) detected within a muscle which can be cast as a MIL problem. Our approach leads to a state-of-the-art solution to the problem of muscle classification. By introducing and analyzing generative models for MIL in a general framework and examining a variety of model structures and components, our work also serves as a methodological guide to modelling MIL tasks. We evaluate our proposed methods both on MUPT datasets and on the MUSK1 dataset, one of the most widely used benchmarks for MIL.

Daniel J. Lizotte, Omid Madani, Russell Greiner

Frequently, acquiring training data has an associated cost. We consider the situation where the learner may purchase data during training, subject TO a budget. IN particular, we examine the CASE WHERE each feature label has an associated cost, AND the total cost OF ALL feature labels acquired during training must NOT exceed the budget.This paper compares methods FOR choosing which feature label TO purchase next, given the budget AND the CURRENT belief state OF naive Bayes model parameters.Whereas active learning has traditionally focused ON myopic(greedy) strategies FOR query selection, this paper presents a tractable method FOR incorporating knowledge OF the budget INTO the decision making process, which improves performance.

Eric B. Laber, Daniel J. Lizotte, Bradley Ferguson

Dynamic treatment regimes operationalize the clinical decision process as a sequence of functions, one for each clinical decision, where each function takes as input up-to-date patient information and gives as output a single recommended treatment. Current methods for estimating optimal dynamic treatment regimes, for example Q-learning, require the specification of a single outcome by which the `goodness' of competing dynamic treatment regimes are measured. However, this is an over-simplification of the goal of clinical decision making, which aims to balance several potentially competing outcomes. For example, often a balance must be struck between treatment effectiveness and side-effect burden. We propose a method for constructing dynamic treatment regimes that accommodates competing outcomes by recommending sets of treatments at each decision point. Formally, we construct a sequence of set-valued functions that take as input up-to-date patient information and give as output a recommended subset of the possible treatments. For a given patient history, the recommended set of treatments contains all treatments that are not inferior according to any of the competing outcomes. When there is more than one decision point, constructing these set-valued functions requires solving a non-trivial enumeration problem. We offer an exact enumeration algorithm by recasting the problem as a linear mixed integer program. The proposed methods are illustrated using data from a depression study and the CATIE schizophrenia study.

Omid Madani, Daniel J. Lizotte, Russell Greiner

Classical learning assumes the learner is given a labeled data sample, from which it learns a model. The field of Active Learning deals with the situation where the learner begins not with a training sample, but instead with resources that it can use to obtain information to help identify the optimal model. To better understand this task, this paper presents and analyses the simplified "(budgeted) active model selection" version, which captures the pure exploration aspect of many active learning problems in a clean and simple problem formulation. Here the learner can use a fixed budget of "model probes" (where each probe evaluates the specified model on a random indistinguishable instance) to identify which of a given set of possible models has the highest expected accuracy. Our goal is a policy that sequentially determines which model to probe next, based on the information observed so far. We present a formal description of this task, and show that it is NPhard in general. We then investigate a number of algorithms for this task, including several existing ones (eg, "Round-Robin", "Interval Estimation", "Gittins") as well as some novel ones (e.g., "Biased-Robin"), describing first their approximation properties and then their empirical performance on various problem instances. We observe empirically that the simple biased-robin algorithm significantly outperforms the other algorithms in the case of identical costs and priors.