The development of neural networks for clinical artificial intelligence (AI) is reliant on interpretability, transparency, and performance. The need to delve into the black-box neural network and derive interpretable explanations of model output is paramount. A task of high clinical importance is predicting the likelihood of a patient being readmitted to hospital in the near future to enable efficient triage. With the increasing adoption of electronic health records (EHRs), there is great interest in applications of natural language processing (NLP) to clinical free-text contained within EHRs. In this work, we apply InfoCal, the current state-of-the-art model that produces extractive rationales for its predictions, to the task of predicting hospital readmission using hospital discharge notes. We compare extractive rationales produced by InfoCal to competitive transformer-based models pretrained on clinical text data and for which the attention mechanism can be used for interpretation. We find each presented model with selected interpretability or feature importance methods yield varying results, with clinical language domain expertise and pretraining critical to performance and subsequent interpretability.
Machine learning (ML) is increasingly often used to inform high-stakes decisions. As complex ML models (e.g., deep neural networks) are often considered black boxes, a wealth of procedures has been developed to shed light on their inner workings and the ways in which their predictions come about, defining the field of 'explainable AI' (XAI). Saliency methods rank input features according to some measure of 'importance'. Such methods are difficult to validate since a formal definition of feature importance is, thus far, lacking. It has been demonstrated that some saliency methods can highlight features that have no statistical association with the prediction target (suppressor variables). To avoid misinterpretations due to such behavior, we propose the actual presence of such an association as a necessary condition and objective preliminary definition for feature importance. We carefully crafted a ground-truth dataset in which all statistical dependencies are well-defined and linear, serving as a benchmark to study the problem of suppressor variables. We evaluate common explanation methods including LRP, DTD, PatternNet, PatternAttribution, LIME, Anchors, SHAP, and permutation-based methods with respect to our objective definition. We show that most of these methods are unable to distinguish important features from suppressors in this setting.
Interpretability is becoming an active research topic as machine learning (ML) models are more widely used to make critical decisions. Tabular data is one of the most commonly used modes of data in diverse applications such as healthcare and finance. Much of the existing interpretability methods used for tabular data only report feature-importance scores -- either locally (per example) or globally (per model) -- but they do not provide interpretation or visualization of how the features interact. We address this limitation by introducing Feature Vectors, a new global interpretability method designed for tabular datasets. In addition to providing feature-importance, Feature Vectors discovers the inherent semantic relationship among features via an intuitive feature visualization technique. Our systematic experiments demonstrate the empirical utility of this new method by applying it to several real-world datasets. We further provide an easy-to-use Python package for Feature Vectors.
Delirium is a common acute onset brain dysfunction in the emergency setting and is associated with higher mortality. It is difficult to detect and monitor since its presentations and risk factors can be different depending on the underlying medical condition of patients. In our study, we aimed to identify subtypes within the delirium population and build subgroup-specific predictive models to detect delirium using Medical Information Mart for Intensive Care IV (MIMIC-IV) data. We showed that clusters exist within the delirium population. Differences in feature importance were also observed for subgroup-specific predictive models. Our work could recalibrate existing delirium prediction models for each delirium subgroup and improve the precision of delirium detection and monitoring for ICU or emergency department patients who had highly heterogeneous medical conditions.
There exist several methods that aim to address the crucial task of understanding the behaviour of AI/ML models. Arguably, the most popular among them are local explanations that focus on investigating model behaviour for individual instances. Several methods have been proposed for local analysis, but relatively lesser effort has gone into understanding if the explanations are robust and accurately reflect the behaviour of underlying models. In this work, we present a survey of the works that analysed the robustness of two classes of local explanations (feature importance and counterfactual explanations) that are popularly used in analysing AI/ML models in finance. The survey aims to unify existing definitions of robustness, introduces a taxonomy to classify different robustness approaches, and discusses some interesting results. Finally, the survey introduces some pointers about extending current robustness analysis approaches so as to identify reliable explainability methods.
We develop various AI models to predict hospitalization on a large (over 110$k$) cohort of COVID-19 positive-tested US patients, sourced from March 2020 to February 2021. Models range from Random Forest to Neural Network (NN) and Time Convolutional NN, where combination of the data modalities (tabular and time dependent) are performed at different stages (early vs. model fusion). Despite high data unbalance, the models reach average precision 0.96-0.98 (0.75-0.85), recall 0.96-0.98 (0.74-0.85), and $F_1$-score 0.97-0.98 (0.79-0.83) on the non-hospitalized (or hospitalized) class. Performances do not significantly drop even when selected lists of features are removed to study model adaptability to different scenarios. However, a systematic study of the SHAP feature importance values for the developed models in the different scenarios shows a large variability across models and use cases. This calls for even more complete studies on several explainability methods before their adoption in high-stakes scenarios.
Feature attributions are a common paradigm for model explanations due to their simplicity in assigning a single numeric score for each input feature to a model. In the actionable recourse setting, wherein the goal of the explanations is to improve outcomes for model consumers, it is often unclear how feature attributions should be correctly used. With this work, we aim to strengthen and clarify the link between actionable recourse and feature attributions. Concretely, we propose a variant of SHAP, Counterfactual SHAP (CF-SHAP), that incorporates counterfactual information to produce a background dataset for use within the marginal (a.k.a. interventional) Shapley value framework. We motivate the need within the actionable recourse setting for careful consideration of background datasets when using Shapley values for feature attributions with numerous synthetic examples. Moreover, we demonstrate the efficacy of CF-SHAP by proposing and justifying a quantitative score for feature attributions, counterfactual-ability, showing that as measured by this metric, CF-SHAP is superior to existing methods when evaluated on public datasets using tree ensembles.
Post-hoc global/local feature attribution methods are progressively being employed to understand the decisions of complex machine learning models. Yet, because of limited amounts of data, it is possible to obtain a diversity of models with good empirical performance but that provide very different explanations for the same prediction, making it hard to derive insight from them. In this work, instead of aiming at reducing the under-specification of model explanations, we fully embrace it and extract logical statements about feature attributions that are consistent across all models with good empirical performance (i.e. all models in the Rashomon Set). We show that partial orders of local/global feature importance arise from this methodology enabling more nuanced interpretations by allowing pairs of features to be incomparable when there is no consensus on their relative importance. We prove that every relation among features present in these partial orders also holds in the rankings provided by existing approaches. Finally, we present three use cases employing hypothesis spaces with tractable Rashomon Sets (Additive models, Kernel Ridge, and Random Forests) and show that partial orders allow one to extract consistent local and global interpretations of models despite their under-specification.
Alzheimer's disease (AD) is a neurodegenerative disease that affects nearly 50 million individuals across the globe and is one of the leading causes of deaths globally. It is projected that by 2050, the number of people affected by the disease would more than double. Consequently, the growing advancements in technology beg the question, can technology be used to predict Alzheimer's for a better and early diagnosis? In this paper, we focus on this very problem. Specifically, we have trained both ML models and neural networks to predict and classify participants based on their speech patterns. We computed a number of linguistic variables using DementiaBank's Pitt Corpus, a database consisting of transcripts of interviews with subjects suffering from multiple neurodegenerative diseases. We then trained both binary classifiers, as well as multiclass classifiers to distinguish AD from normal aging and other neurodegenerative diseases. We also worked on establishing the link between specific speech factors that can help determine the onset of AD. Confusion matrices and feature importance graphs have been plotted model-wise to compare the performances of our models. In both multiclass and binary classification, neural networks were found to outperform the other models with a testing accuracy of 76.44% and 92.05% respectively. For the feature importance, it was concluded that '%_PRESP' (present participle), '%_3S' (3rd person present tense markers) were two of the most important speech features for our classifiers in predicting AD.
Background. With the rise of highly portable, wireless, and low-cost ultrasound devices and automatic ultrasound acquisition techniques, an automated interpretation method requiring only a limited set of views as input could make preliminary cardiovascular disease diagnoses more accessible. In this study, we developed a deep learning (DL) method for automated detection of impaired left ventricular (LV) function and aortic valve (AV) regurgitation from apical four-chamber (A4C) ultrasound cineloops and investigated which anatomical structures or temporal frames provided the most relevant information for the DL model to enable disease classification. Methods and Results. A4C ultrasounds were extracted from 3,554 echocardiograms of patients with either impaired LV function (n=928), AV regurgitation (n=738), or no significant abnormalities (n=1,888). Two convolutional neural networks (CNNs) were trained separately to classify the respective disease cases against normal cases. The overall classification accuracy of the impaired LV function detection model was 86%, and that of the AV regurgitation detection model was 83%. Feature importance analyses demonstrated that the LV myocardium and mitral valve were important for detecting impaired LV function, while the tip of the mitral valve anterior leaflet, during opening, was considered important for detecting AV regurgitation. Conclusion. The proposed method demonstrated the feasibility of a 3D CNN approach in detection of impaired LV function and AV regurgitation using A4C ultrasound cineloops. The current research shows that DL methods can exploit large training data to detect diseases in a different way than conventionally agreed upon methods, and potentially reveal unforeseen diagnostic image features.
With the widespread use of machine learning to support decision-making, it is increasingly important to verify and understand the reasons why a particular output is produced. Although post-training feature importance approaches assist this interpretation, there is an overall lack of consensus regarding how feature importance should be quantified, making explanations of model predictions unreliable. In addition, many of these explanations depend on the specific machine learning approach employed and on the subset of data used when calculating feature importance. A possible solution to improve the reliability of explanations is to combine results from multiple feature importance quantifiers from different machine learning approaches coupled with re-sampling. Current state-of-the-art ensemble feature importance fusion uses crisp techniques to fuse results from different approaches. There is, however, significant loss of information as these approaches are not context-aware and reduce several quantifiers to a single crisp output. More importantly, their representation of 'importance' as coefficients is misleading and incomprehensible to end-users and decision makers. Here we show how the use of fuzzy data fusion methods can overcome some of the important limitations of crisp fusion methods.
The challenge of solving data mining problems in e-commerce applications such as recommendation system (RS) and click-through rate (CTR) prediction is how to make inferences by constructing combinatorial features from a large number of categorical features while preserving the interpretability of the method. In this paper, we propose Automatic Embedded Feature Engineering(AEFE), an automatic feature engineering framework for representing categorical features, which consists of various components including custom paradigm feature construction and multiple feature selection. By selecting the potential field pairs intelligently and generating a series of interpretable combinatorial features, our framework can provide a set of unseen generated features for enhancing model performance and then assist data analysts in discovering the feature importance for particular data mining tasks. Furthermore, AEFE is distributed implemented by task-parallelism, data sampling, and searching schema based on Matrix Factorization field combination, to optimize the performance and enhance the efficiency and scalability of the framework. Experiments conducted on some typical e-commerce datasets indicate that our method outperforms the classical machine learning models and state-of-the-art deep learning models.
Numerous algorithms have been proposed for detecting anomalies (outliers, novelties) in an unsupervised manner. Unfortunately, it is not trivial, in general, to understand why a given sample (record) is labelled as an anomaly and thus diagnose its root causes. We propose the following reduced-dimensionality, surrogate model approach to explain detector decisions: approximate the detection model with another one that employs only a small subset of features. Subsequently, samples can be visualized in this low-dimensionality space for human understanding. To this end, we develop PROTEUS, an AutoML pipeline to produce the surrogate model, specifically designed for feature selection on imbalanced datasets. The PROTEUS surrogate model can not only explain the training data, but also the out-of-sample (unseen) data. In other words, PROTEUS produces predictive explanations by approximating the decision surface of an unsupervised detector. PROTEUS is designed to return an accurate estimate of out-of-sample predictive performance to serve as a metric of the quality of the approximation. Computational experiments confirm the efficacy of PROTEUS to produce predictive explanations for different families of detectors and to reliably estimate their predictive performance in unseen data. Unlike several ad-hoc feature importance methods, PROTEUS is robust to high-dimensional data.
Recent developments in graph neural networks show promise for predicting the occurrence of abnormal grain growth, which has been a particularly challenging area of research due to its apparent stochastic nature. In this study, we generate a large dataset of Monte Carlo simulations of abnormal grain growth. We train simple graph convolution networks to predict which initial microstructures will exhibit abnormal grain growth, and compare the results to a standard computer vision approach for the same task. The graph neural network outperformed the computer vision method and achieved 73% prediction accuracy and fewer false positives. It also provided some physical insight into feature importance and the relevant length scale required to maximize predictive performance. Analysis of the uncertainty in the Monte Carlo simulations provides additional insights for ongoing work in this area.
In recent years, Artificial Intelligence (AI) algorithms have been proven to outperform traditional statistical methods in terms of predictivity, especially when a large amount of data was available. Nevertheless, the "black box" nature of AI models is often a limit for a reliable application in high-stakes fields like diagnostic techniques, autonomous guide, etc. Recent works have shown that an adequate level of interpretability could enforce the more general concept of model trustworthiness. The basic idea of this paper is to exploit the human prior knowledge of the features' importance for a specific task, in order to coherently aid the phase of the model's fitting. This sort of "weighted" AI is obtained by extending the empirical loss with a regularization term encouraging the importance of the features to follow predetermined constraints. This procedure relies on local methods for the feature importance computation, e.g. LRP, LIME, etc. that are the link between the model weights to be optimized and the user-defined constraints on feature importance. In the fairness area, promising experimental results have been obtained for the Adult dataset. Many other possible applications of this model agnostic theoretical framework are described.
Amongst others, the adoption of Rectified Linear Units (ReLUs) is regarded as one of the ingredients of the success of deep learning. ReLU activation has been shown to mitigate the vanishing gradient issue, to encourage sparsity in the learned parameters, and to allow for efficient backpropagation. In this paper, we recognize that the non-linear behavior of the ReLU function gives rise to a natural clustering when the pattern of active neurons is considered. This observation helps to deepen the learning mechanism of the network; in fact, we demonstrate that, within each cluster, the network can be fully represented as an affine map. The consequence is that we are able to recover an explanation, in the form of feature importance, for the predictions done by the network to the instances belonging to the cluster. Therefore, the methodology we propose is able to increase the level of interpretability of a fully connected feedforward ReLU neural network, downstream from the fitting phase of the model, without altering the structure of the network. A simulation study and the empirical application to the Titanic dataset, show the capability of the method to bridge the gap between the algorithm optimization and the human understandability of the black box deep ReLU networks.
Electrocorticogram (ECoG) well characterizes hand movement intentions and gestures. In the present work we aim to investigate the possibility to enhance hand pose classification, in a Rock-Paper-Scissor - and Rest - task, by introducing topological descriptors of time series data. We hypothesized that an innovative approach based on topological data analysis can extract hidden information that are not detectable with standard Brain Computer Interface (BCI)techniques. To investigate this hypothesis, we integrate topological features together with power band features and feed them to several standard classifiers, e.g. Random Forest,Gradient Boosting. Model selection is thus completed after a meticulous phase of bayesian hyperparameter optimization. With our method, we observed robust results in terms of ac-curacy for a four-labels classification problem, with limited available data. Through feature importance investigation, we conclude that topological descriptors are able to extract useful discriminative information and provide novel insights.Since our data are restricted to single-patient recordings, generalization might be limited. Nevertheless, our method can be extended and applied to a wide range of neurophysiological recordings and it might be an intriguing point of departure for future studies.
The recent increase in the deployment of machine learning models in critical domains such as healthcare, criminal justice, and finance has highlighted the need for trustworthy methods that can explain these models to stakeholders. Feature importance methods (e.g. gain and SHAP) are among the most popular explainability methods used to address this need. For any explainability technique to be trustworthy and meaningful, it has to provide an explanation that is accurate and stable. Although the stability of local feature importance methods (explaining individual predictions) has been studied before, there is yet a knowledge gap about the stability of global features importance methods (explanations for the whole model). Additionally, there is no study that evaluates and compares the accuracy of global feature importance methods with respect to feature ordering. In this paper, we evaluate the accuracy and stability of global feature importance methods through comprehensive experiments done on simulations as well as four real-world datasets. We focus on tree-based ensemble methods as they are used widely in industry and measure the accuracy and stability of explanations under two scenarios: 1) when inputs are perturbed 2) when models are perturbed. Our findings provide a comparison of these methods under a variety of settings and shed light on the limitations of global feature importance methods by indicating their lack of accuracy with and without noisy inputs, as well as their lack of stability with respect to: 1) increase in input dimension or noise in the data; 2) perturbations in models initialized by different random seeds or hyperparameter settings.
In artificial neural networks, understanding the contributions of input features on the prediction fosters model explainability and delivers relevant information about the dataset. While typical setups for feature importance ranking assess input features individually, in this study, we go one step further and rank the importance of groups of features, denoted as feature-blocks. A feature-block can contain features of a specific type or features derived from a particular source, which are presented to the neural network in separate input branches (multiblock ANNs). This work presents three methods pursuing distinct strategies to rank features in multiblock ANNs by their importance: (1) a composite strategy building on individual feature importance rankings, (2) a knock-in, and (3) a knock-out strategy. While the composite strategy builds on state-of-the-art feature importance rankings, knock-in and knock-out strategies evaluate the block as a whole via a mutual information criterion. Our experiments consist of a simulation study validating all three approaches, followed by a case study on two distinct real-world datasets to compare the strategies. We conclude that each strategy has its merits for specific application scenarios.
Traditional hand-crafted linguistically-informed features have often been used for distinguishing between translated and original non-translated texts. By contrast, to date, neural architectures without manual feature engineering have been less explored for this task. In this work, we (i) compare the traditional feature-engineering-based approach to the feature-learning-based one and (ii) analyse the neural architectures in order to investigate how well the hand-crafted features explain the variance in the neural models' predictions. We use pre-trained neural word embeddings, as well as several end-to-end neural architectures in both monolingual and multilingual settings and compare them to feature-engineering-based SVM classifiers. We show that (i) neural architectures outperform other approaches by more than 20 accuracy points, with the BERT-based model performing the best in both the monolingual and multilingual settings; (ii) while many individual hand-crafted translationese features correlate with neural model predictions, feature importance analysis shows that the most important features for neural and classical architectures differ; and (iii) our multilingual experiments provide empirical evidence for translationese universals across languages.