Sofie Van Hoecke

Jef Jonkers, Glenn Van Wallendael, Luc Duchateau et al.

Proper scoring rules provide a rigorous theoretical basis for the training and evaluation of probabilistic forecasts. However, in the presence of right censoring, the event time is only partially observed, rendering conventional scoring rules inapplicable in their standard form. We propose a framework for proper scoring of right-censored survival outcomes based on a simple idea: first, map the predictive distribution through the censoring mechanism, then apply the underlying proper score on the induced observed-data law. This yields localized scores for fixed censoring times and marginalized scores when the censoring time is random or only partially observed. The resulting construction recovers familiar right-censored likelihood and IPCW-type criteria within a coherent framework, while also yielding right-censored versions of the CRPS, pinball loss, Brier score, and energy score. We show that the marginalized score is proper under conditional independent censoring and strictly proper on the identifiable region. The same principle also leads to censored engression, a sample-based learning objective for multivariate right-censored survival modeling. In experiments, our scores correctly rank the oracle forecast across several censoring regimes, whereas forecast-dependent plug-in weighted scores can exhibit ranking reversals. Censored engression likewise substantially improves over naive training on censored outcomes.

Jarne Verhaeghe, Jeroen Van Der Donckt, Femke Ongenae et al.

Feature selection is a crucial step in developing robust and powerful machine learning models. Feature selection techniques can be divided into two categories: filter and wrapper methods. While wrapper methods commonly result in strong predictive performances, they suffer from a large computational complexity and therefore take a significant amount of time to complete, especially when dealing with high-dimensional feature sets. Alternatively, filter methods are considerably faster, but suffer from several other disadvantages, such as (i) requiring a threshold value, (ii) not taking into account intercorrelation between features, and (iii) ignoring feature interactions with the model. To this end, we present powershap, a novel wrapper feature selection method, which leverages statistical hypothesis testing and power calculations in combination with Shapley values for quick and intuitive feature selection. Powershap is built on the core assumption that an informative feature will have a larger impact on the prediction compared to a known random feature. Benchmarks and simulations show that powershap outperforms other filter methods with predictive performances on par with wrapper methods while being significantly faster, often even reaching half or a third of the execution time. As such, powershap provides a competitive and quick algorithm that can be used by various models in different domains. Furthermore, powershap is implemented as a plug-and-play and open-source sklearn component, enabling easy integration in conventional data science pipelines. User experience is even further enhanced by also providing an automatic mode that automatically tunes the hyper-parameters of the powershap algorithm, allowing to use the algorithm without any configuration needed.

Jonas Van Der Donckt, Jeroen Van Der Donckt, Emiel Deprost et al.

Visual analytics is arguably the most important step in getting acquainted with your data. This is especially the case for time series, as this data type is hard to describe and cannot be fully understood when using for example summary statistics. To realize effective time series visualization, four requirements have to be met; a tool should be (1) interactive, (2) scalable to millions of data points, (3) integrable in conventional data science environments, and (4) highly configurable. We observe that open source Python visualization toolkits empower data scientists in most visual analytics tasks, but lack the combination of scalability and interactivity to realize effective time series visualization. As a means to facilitate these requirements, we created Plotly-Resampler, an open source Python library. Plotly-Resampler is an add-on for Plotly's Python bindings, enhancing line chart scalability on top of an interactive toolkit by aggregating the underlying data depending on the current graph view. Plotly-Resampler is built to be snappy, as the reactivity of a tool qualitatively affects how analysts visually explore and analyze data. A benchmark task highlights how our toolkit scales better than alternatives in terms of number of samples and time series. Additionally, Plotly-Resampler's flexible data aggregation functionality paves the path towards researching novel aggregation techniques. Plotly-Resampler's integrability, together with its configurability, convenience, and high scalability, allows to effectively analyze high-frequency data in your day-to-day Python environment.

Jeroen Van Der Donckt, Jonas Van Der Donckt, Emiel Deprost et al.

Over the last few years, research in automatic sleep scoring has mainly focused on developing increasingly complex deep learning architectures. However, recently these approaches achieved only marginal improvements, often at the expense of requiring more data and more expensive training procedures. Despite all these efforts and their satisfactory performance, automatic sleep staging solutions are not widely adopted in a clinical context yet. We argue that most deep learning solutions for sleep scoring are limited in their real-world applicability as they are hard to train, deploy, and reproduce. Moreover, these solutions lack interpretability and transparency, which are often key to increase adoption rates. In this work, we revisit the problem of sleep stage classification using classical machine learning. Results show that competitive performance can be achieved with a conventional machine learning pipeline consisting of preprocessing, feature extraction, and a simple machine learning model. In particular, we analyze the performance of a linear model and a non-linear (gradient boosting) model. Our approach surpasses state-of-the-art (that uses the same data) on two public datasets: Sleep-EDF SC-20 (MF1 0.810) and Sleep-EDF ST (MF1 0.795), while achieving competitive results on Sleep-EDF SC-78 (MF1 0.775) and MASS SS3 (MF1 0.817). We show that, for the sleep stage scoring task, the expressiveness of an engineered feature vector is on par with the internally learned representations of deep learning models. This observation opens the door to clinical adoption, as a representative feature vector allows to leverage both the interpretability and successful track record of traditional machine learning models.

Vic Degraeve, Gilles Vandewiele, Femke Ongenae et al.

The inception of the Relational Graph Convolutional Network (R-GCN) marked a milestone in the Semantic Web domain as a widely cited method that generalises end-to-end hierarchical representation learning to Knowledge Graphs (KGs). R-GCNs generate representations for nodes of interest by repeatedly aggregating parameterised, relation-specific transformations of their neighbours. However, in this paper, we argue that the the R-GCN's main contribution lies in this "message passing" paradigm, rather than the learned weights. To this end, we introduce the "Random Relational Graph Convolutional Network" (RR-GCN), which leaves all parameters untrained and thus constructs node embeddings by aggregating randomly transformed random representations from neighbours, i.e., with no learned parameters. We empirically show that RR-GCNs can compete with fully trained R-GCNs in both node classification and link prediction settings.

Jarne Verhaeghe, Jef Jonkers, Sofie Van Hoecke

Understanding the dose-response relation between a continuous treatment and the outcome for an individual can greatly drive decision-making, particularly in areas like personalized drug dosing and personalized healthcare interventions. Point estimates are often insufficient in these high-risk environments, highlighting the need for uncertainty quantification to support informed decisions. Conformal prediction, a distribution-free and model-agnostic method for uncertainty quantification, has seen limited application in continuous treatments or dose-response models. To address this gap, we propose a novel methodology that frames the causal dose-response problem as a covariate shift, leveraging weighted conformal prediction. By incorporating propensity estimation, conformal predictive systems, and likelihood ratios, we present a practical solution for generating prediction intervals for dose-response models. Additionally, our method approximates local coverage for every treatment value by applying kernel functions as weights in weighted conformal prediction. Finally, we use a new synthetic benchmark dataset to demonstrate the significance of covariate shift assumptions in achieving robust prediction intervals for dose-response models.

Sandeep Ramachandra, Gilles Vandewiele, David Vander Mijnsbrugge et al.

A paper of Alsinglawi et al was recently accepted and published in Scientific Reports. In this paper, the authors aim to predict length of stay (LOS), discretized into either long (> 7 days) or short stays (< 7 days), of lung cancer patients in an ICU department using various machine learning techniques. The authors claim to achieve perfect results with an Area Under the Receiver Operating Characteristic curve (AUROC) of 100% with a Random Forest (RF) classifier with ADASYN class balancing over sampling technique, which if accurate could have significant implications for hospital management. However, we have identified several methodological flaws within the manuscript which cause the results to be overly optimistic and would have serious consequences if used in a clinical practice. Moreover, the reporting of the methodology is unclear and many important details are missing from the manuscript, which makes reproduction extremely difficult. We highlight the effect these oversights have had on the result and provide a more believable result of 88.91% AUROC when these oversights are corrected.

Jeroen Van Der Donckt, Jonas Van Der Donckt, Sofie Van Hoecke

This work presents the solution of the Signal Sleuths team for the 2024 SHL recognition challenge. The challenge involves detecting transportation modes using shuffled, non-overlapping 5-second windows of phone movement data, with exactly one of the three available modalities (accelerometer, gyroscope, magnetometer) randomly missing. Data analysis indicated a significant distribution shift between train and validation data, necessitating a magnitude and rotation-invariant approach. We utilize traditional machine learning, focusing on robust processing, feature extraction, and rotation-invariant aggregation. An ablation study showed that relying solely on the frequently used signal magnitude vector results in the poorest performance. Conversely, our proposed rotation-invariant aggregation demonstrated substantial improvement over using rotation-aware features, while also reducing the feature vector length. Moreover, z-normalization proved crucial for creating robust spectral features.

Jonas Van Der Donckt, Jeroen Van Der Donckt, Sofie Van Hoecke

This work presents the solution of the Signal Sleuths team for the 2024 HASCA WEAR challenge. The challenge focuses on detecting 18 workout activities (and the null class) using accelerometer data from 4 wearables - one worn on each limb. Data analysis revealed inconsistencies in wearable orientation within and across participants, leading to exploring novel multi-wearable data augmentation techniques. We investigate three models using a fixed feature set: (i) "raw": using all data as is, (ii) "left-right swapping": augmenting data by swapping left and right limb pairs, and (iii) "upper-lower limb paring": stacking data by using upper-lower limb pair combinations (2 wearables). Our experiments utilize traditional machine learning with multi-window feature extraction and temporal smoothing. Using 3-fold cross-validation, the raw model achieves a macro F1-score of 90.01%, whereas left-right swapping and upper-lower limb paring improve the scores to 91.30% and 91.87% respectively.

Jef Jonkers, Glenn Van Wallendael, Luc Duchateau et al.

Conformal Predictive Systems (CPS) offer a versatile framework for constructing predictive distributions, allowing for calibrated inference and informative decision-making. However, their applicability has been limited to scenarios adhering to the Independent and Identically Distributed (IID) model assumption. This paper extends CPS to accommodate scenarios characterized by covariate shifts. We therefore propose Weighted CPS (WCPS), akin to Weighted Conformal Prediction (WCP), leveraging likelihood ratios between training and testing covariate distributions. This extension enables the construction of nonparametric predictive distributions capable of handling covariate shifts. We present theoretical underpinnings and conjectures regarding the validity and efficacy of WCPS and demonstrate its utility through empirical evaluations on both synthetic and real-world datasets. Our simulation experiments indicate that WCPS are probabilistically calibrated under covariate shift.

Jef Jonkers, Jarne Verhaeghe, Glenn Van Wallendael et al.

Generating probabilistic forecasts of potential outcomes and individual treatment effects (ITE) is essential for risk-aware decision-making in domains such as healthcare, policy, marketing, and finance. We propose two novel methods: the conformal convolution T-learner (CCT) and the conformal Monte Carlo (CMC) meta-learner, that generate full predictive distributions of both potential outcomes and ITEs. Our approaches combine weighted conformal predictive systems with either analytic convolution of potential outcome distributions or Monte Carlo sampling, addressing covariate shift through propensity score weighting. In contrast to other approaches that allow the generation of potential outcome predictive distributions, our approaches are model agnostic, universal, and come with finite-sample guarantees of probabilistic calibration under knowledge of the propensity score. Regarding estimating the ITE distribution, we formally characterize how assumptions about potential outcomes' noise dependency impact distribution validity and establish universal consistency under independence noise assumptions. Experiments on synthetic and semi-synthetic datasets demonstrate that the proposed methods achieve probabilistically calibrated predictive distributions while maintaining narrow prediction intervals and having performant continuous ranked probability scores. Besides probabilistic forecasting performance, we observe significant efficiency gains for the CCT- and CMC meta-learners compared to other conformal approaches that produce prediction intervals for ITE with coverage guarantees.

Jef Jonkers, Luc Duchateau, Glenn Van Wallendael et al.

Anatomical landmark localization in 2D/3D images is a critical task in medical imaging. Although many general-purpose tools exist for landmark localization in classical computer vision tasks, such as pose estimation, they lack the specialized features and modularity necessary for anatomical landmark localization applications in the medical domain. Therefore, we introduce landmarker, a Python package built on PyTorch. The package provides a comprehensive, flexible toolkit for developing and evaluating landmark localization algorithms, supporting a range of methodologies, including static and adaptive heatmap regression. landmarker enhances the accuracy of landmark identification, streamlines research and development processes, and supports various image formats and preprocessing pipelines. Its modular design allows users to customize and extend the toolkit for specific datasets and applications, accelerating innovation in medical imaging. landmarker addresses a critical need for precision and customization in landmark localization tasks not adequately met by existing general-purpose pose estimation tools.

Jef Jonkers, Frank Coopman, Luc Duchateau et al.

Automatic anatomical landmark localization in medical imaging requires not just accurate predictions but reliable uncertainty quantification for effective clinical decision support. Current uncertainty quantification approaches often fall short, particularly when combined with normality assumptions, systematically underestimating total predictive uncertainty. This paper introduces conformal prediction as a framework for reliable uncertainty quantification in anatomical landmark localization, addressing a critical gap in automatic landmark localization. We present two novel approaches guaranteeing finite-sample validity for multi-output prediction: Multi-output Regression-as-Classification Conformal Prediction (M-R2CCP) and its variant Multi-output Regression to Classification Conformal Prediction set to Region (M-R2C2R). Unlike conventional methods that produce axis-aligned hyperrectangular or ellipsoidal regions, our approaches generate flexible, non-convex prediction regions that better capture the underlying uncertainty structure of landmark predictions. Through extensive empirical evaluation across multiple 2D and 3D datasets, we demonstrate that our methods consistently outperform existing multi-output conformal prediction approaches in both validity and efficiency. This work represents a significant advancement in reliable uncertainty estimation for anatomical landmark localization, providing clinicians with trustworthy confidence measures for their diagnoses. While developed for medical imaging, these methods show promise for broader applications in multi-output regression problems.

Nathan Vandemoortele, Bram Steenwinckel, Femke Ongenae et al.

We present Label Space Reduction (LSR), a novel method for improving zero-shot classification performance of Large Language Models (LLMs). LSR iteratively refines the classification label space by systematically ranking and reducing candidate classes, enabling the model to concentrate on the most relevant options. By leveraging unlabeled data with the statistical learning capabilities of data-driven models, LSR dynamically optimizes the label space representation at test time. Our experiments across seven benchmarks demonstrate that LSR improves macro-F1 scores by an average of 7.0% (up to 14.2%) with Llama-3.1-70B and 3.3% (up to 11.1%) with Claude-3.5-Sonnet compared to standard zero-shot classification baselines. To reduce the computational overhead of LSR, which requires an additional LLM call at each iteration, we propose distilling the model into a probabilistic classifier, allowing for efficient inference.

Jonas Van Der Donckt, Jeroen Van Der Donckt, Emiel Deprost et al.

Time series processing and feature extraction are crucial and time-intensive steps in conventional machine learning pipelines. Existing packages are limited in their applicability, as they cannot cope with irregularly-sampled or asynchronous data and make strong assumptions about the data format. Moreover, these packages do not focus on execution speed and memory efficiency, resulting in considerable overhead. We present $\texttt{tsflex}$, a Python toolkit for time series processing and feature extraction, that focuses on performance and flexibility, enabling broad applicability. This toolkit leverages window-stride arguments of the same data type as the sequence-index, and maintains the sequence-index through all operations. $\texttt{tsflex}$ is flexible as it supports (1) multivariate time series, (2) multiple window-stride configurations, and (3) integrates with processing and feature functions from other packages, while (4) making no assumptions about the data sampling regularity, series alignment, and data type. Other functionalities include multiprocessing, detailed execution logging, chunking sequences, and serialization. Benchmarks show that $\texttt{tsflex}$ is faster and more memory-efficient compared to similar packages, while being more permissive and flexible in its utilization.

Gilles Vandewiele, Isabelle Dehaene, György Kovács et al.

Information extracted from electrohysterography recordings could potentially prove to be an interesting additional source of information to estimate the risk on preterm birth. Recently, a large number of studies have reported near-perfect results to distinguish between recordings of patients that will deliver term or preterm using a public resource, called the Term/Preterm Electrohysterogram database. However, we argue that these results are overly optimistic due to a methodological flaw being made. In this work, we focus on one specific type of methodological flaw: applying over-sampling before partitioning the data into mutually exclusive training and testing sets. We show how this causes the results to be biased using two artificial datasets and reproduce results of studies in which this flaw was identified. Moreover, we evaluate the actual impact of over-sampling on predictive performance, when applied prior to data partitioning, using the same methodologies of related studies, to provide a realistic view of these methodologies' generalization capabilities. We make our research reproducible by providing all the code under an open license.

Mijung Kim, Olivier Janssens, Ho-min Park et al.

Glaucoma is a major eye disease, leading to vision loss in the absence of proper medical treatment. Current diagnosis of glaucoma is performed by ophthalmologists who are often analyzing several types of medical images generated by different types of medical equipment. Capturing and analyzing these medical images is labor-intensive and expensive. In this paper, we present a novel computational approach towards glaucoma diagnosis and localization, only making use of eye fundus images that are analyzed by state-of-the-art deep learning techniques. Specifically, our approach leverages Convolutional Neural Networks (CNNs) and Gradient-weighted Class Activation Mapping (Grad-CAM) for glaucoma diagnosis and localization, respectively. Quantitative and qualitative results, as obtained for a small-sized dataset with no segmentation ground truth, demonstrate that the proposed approach is promising, for instance achieving an accuracy of 0.91$\pm0.02$ and an ROC-AUC score of 0.94 for the diagnosis task. Furthermore, we present a publicly available prototype web application that integrates our predictive model, with the goal of making effective glaucoma diagnosis available to a wide audience.

Gilles Vandewiele, Olivier Janssens, Femke Ongenae et al.

Models obtained by decision tree induction techniques excel in being interpretable.However, they can be prone to overfitting, which results in a low predictive performance. Ensemble techniques are able to achieve a higher accuracy. However, this comes at a cost of losing interpretability of the resulting model. This makes ensemble techniques impractical in applications where decision support, instead of decision making, is crucial. To bridge this gap, we present the GENESIM algorithm that transforms an ensemble of decision trees to a single decision tree with an enhanced predictive performance by using a genetic algorithm. We compared GENESIM to prevalent decision tree induction and ensemble techniques using twelve publicly available data sets. The results show that GENESIM achieves a better predictive performance on most of these data sets than decision tree induction techniques and a predictive performance in the same order of magnitude as the ensemble techniques. Moreover, the resulting model of GENESIM has a very low complexity, making it very interpretable, in contrast to ensemble techniques.