Filters








23 Hits in 0.95 sec

Causal Simulations for Uplift Modeling [article]

Jeroen Berrevoets, Wouter Verbeke
2019 arXiv   pre-print
Uplift modeling requires experimental data, preferably collected in random fashion. This places a logistical and financial burden upon any organisation aspiring such models. Once deployed, uplift models are subject to effects from concept drift. Hence, methods are being developed that are able to learn from newly gained experience, as well as handle drifting environments. As these new methods attempt to eliminate the need for experimental data, another approach to test such methods must be
more » ... lated. Therefore, we propose a method to simulate environments that offer causal relationships in their parameters.
arXiv:1902.00287v1 fatcat:zfi6bax5iffbhb5uwgiqngltwm

Optimising Individual-Treatment-Effect Using Bandits [article]

Jeroen Berrevoets, Sam Verboven, Wouter Verbeke
2019 arXiv   pre-print
Applying causal inference models in areas such as economics, healthcare and marketing receives great interest from the machine learning community. In particular, estimating the individual-treatment-effect (ITE) in settings such as precision medicine and targeted advertising has peaked in application. Optimising this ITE under the strong-ignorability-assumption -- meaning all confounders expressing influence on the outcome of a treatment are registered in the data -- is often referred to as
more » ... t modeling (UM). While these techniques have proven useful in many settings, they suffer vividly in a dynamic environment due to concept drift. Take for example the negative influence on a marketing campaign when a competitor product is released. To counter this, we propose the uplifted contextual multi-armed bandit (U-CMAB), a novel approach to optimise the ITE by drawing upon bandit literature. Experiments on real and simulated data indicate that our proposed approach compares favourably against the state-of-the-art. All our code can be found online at https://github.com/vub-dl/u-cmab.
arXiv:1910.07265v1 fatcat:s2w2pqv55beqrp6rvtvwbsla64

Autoencoders for strategic decision support [article]

Sam Verboven, Jeroen Berrevoets, Chris Wuytens, Bart Baesens, Wouter Verbeke
2020 arXiv   pre-print
In the majority of executive domains, a notion of normality is involved in most strategic decisions. However, few data-driven tools that support strategic decision-making are available. We introduce and extend the use of autoencoders to provide strategically relevant granular feedback. A first experiment indicates that experts are inconsistent in their decision making, highlighting the need for strategic decision support. Furthermore, using two large industry-provided human resources datasets,
more » ... he proposed solution is evaluated in terms of ranking accuracy, synergy with human experts, and dimension-level feedback. This three-point scheme is validated using (a) synthetic data, (b) the perspective of data quality, (c) blind expert validation, and (d) transparent expert evaluation. Our study confirms several principal weaknesses of human decision-making and stresses the importance of synergy between a model and humans. Moreover, unsupervised learning and in particular the autoencoder are shown to be valuable tools for strategic decision-making.
arXiv:2005.01075v1 fatcat:burtdpmjo5gpjjjygaqgrrjsuu

Disentangled Counterfactual Recurrent Networks for Treatment Effect Inference over Time [article]

Jeroen Berrevoets, Alicia Curth, Ioana Bica, Eoin McKinney, Mihaela van der Schaar
2021 arXiv   pre-print
Hill [2011] , Johansson et al. [2016] , Shalit et al. [2017] , Alaa and van der Schaar [2018] , Yoon et al. [2018] , Hassanpour and Greiner [2020] , Zhang et al. [2020] , Berrevoets et al. [2020  ... 
arXiv:2112.03811v1 fatcat:gwu3lwz6rzcwxkcvmuzhfmil4i

Combining Observational and Randomized Data for Estimating Heterogeneous Treatment Effects [article]

Tobias Hatt, Jeroen Berrevoets, Alicia Curth, Stefan Feuerriegel, Mihaela van der Schaar
2022 arXiv   pre-print
., Berrevoets et al., 2020 Berrevoets et al., , 2021;; Hatt and Feuerriegel, 2021a) . This allows us to learn the bias more efficiently by exploiting structural similarities.  ... 
arXiv:2202.12891v1 fatcat:mtmnxkpxibaf5hi65upwtmr6pm

HydaLearn: Highly Dynamic Task Weighting for Multi-task Learning with Auxiliary Tasks [article]

Sam Verboven, Muhammad Hafeez Chaudhary, Jeroen Berrevoets, Wouter Verbeke
2020 arXiv   pre-print
Multi-task learning (MTL) can improve performance on a task by sharing representations with one or more related auxiliary-tasks. Usually, MTL-networks are trained on a composite loss function formed by a constant weighted combination of the separate task losses. In practice, constant loss weights lead to poor results for two reasons: (i) the relevance of the auxiliary tasks can gradually drift throughout the learning process; (ii) for mini-batch based optimisation, the optimal task weights vary
more » ... significantly from one update to the next depending on mini-batch sample composition. We introduce HydaLearn, an intelligent weighting algorithm that connects main-task gain to the individual task gradients, in order to inform dynamic loss weighting at the mini-batch level, addressing i and ii. Using HydaLearn, we report performance increases on synthetic data, as well as on two supervised learning domains.
arXiv:2008.11643v1 fatcat:ilffciybh5bxtjuitaojmpvv2e

Why you should stop predicting customer churn and start using uplift models

Floris Devriendt, Jeroen Berrevoets, Wouter Verbeke
2019 Information Sciences  
doi:10.1016/j.ins.2019.12.075 fatcat:tv4v4cggkfgplg6dt27m2k43im

DECAF: Generating Fair Synthetic Data Using Causally-Aware Generative Networks [article]

Boris van Breugel, Trent Kyono, Jeroen Berrevoets, Mihaela van der Schaar
2021 arXiv   pre-print
Machine learning models have been criticized for reflecting unfair biases in the training data. Instead of solving for this by introducing fair learning algorithms directly, we focus on generating fair synthetic data, such that any downstream learner is fair. Generating fair synthetic data from unfair data - while remaining truthful to the underlying data-generating process (DGP) - is non-trivial. In this paper, we introduce DECAF: a GAN-based fair synthetic data generator for tabular data.
more » ... DECAF we embed the DGP explicitly as a structural causal model in the input layers of the generator, allowing each variable to be reconstructed conditioned on its causal parents. This procedure enables inference time debiasing, where biased edges can be strategically removed for satisfying user-defined fairness requirements. The DECAF framework is versatile and compatible with several popular definitions of fairness. In our experiments, we show that DECAF successfully removes undesired bias and - in contrast to existing methods - is capable of generating high-quality synthetic data. Furthermore, we provide theoretical guarantees on the generator's convergence and the fairness of downstream models.
arXiv:2110.12884v2 fatcat:hzksxlullbganns2lk7ufhh24q

To Impute or not to Impute? Missing Data in Treatment Effect Estimation [article]

Jeroen Berrevoets, Fergus Imrie, Trent Kyono, James Jordon, Mihaela van der Schaar
2022 arXiv   pre-print
Missing data is a systemic problem in practical scenarios that causes noise and bias when estimating treatment effects. This makes treatment effect estimation from data with missingness a particularly tricky endeavour. A key reason for this is that standard assumptions on missingness are rendered insufficient due to the presence of an additional variable, treatment, besides the individual and the outcome. Having a treatment variable introduces additional complexity with respect to why some
more » ... bles are missing that is not fully explored by previous work. In our work we identify a new missingness mechanism, which we term mixed confounded missingness (MCM), where some missingness determines treatment selection and other missingness is determined by treatment selection. Given MCM, we show that naively imputing all data leads to poor performing treatment effects models, as the act of imputation effectively removes information necessary to provide unbiased estimates. However, no imputation at all also leads to biased estimates, as missingness determined by treatment divides the population in distinct subpopulations, where estimates across these populations will be biased. Our solution is selective imputation, where we use insights from MCM to inform precisely which variables should be imputed and which should not. We empirically demonstrate how various learners benefit from selective imputation compared to other solutions for missing data.
arXiv:2202.02096v2 fatcat:5ycrak3sgneankbzorwuury6qe

Identifiable Energy-based Representations: An Application to Estimating Heterogeneous Causal Effects [article]

Yao Zhang and Jeroen Berrevoets and Mihaela van der Schaar
2022 arXiv   pre-print
Nabi and Shpitser [2020] and Berrevoets et al. [2020] propose methods to deal with high-dimensional treatment variables, which is not the problem considered in our paper. Covariates selection.  ... 
arXiv:2108.03039v3 fatcat:c726k7f3z5dddobcw2uspknshq

Quality of outpatient parenteral antimicrobial therapy (OPAT) care from the patient's perspective: a qualitative study

Marvin A H Berrevoets, Anke J M Oerlemans, Mirjam Tromp, Bart Jan Kullberg, Jaap Ten Oever, Jeroen A Schouten, Marlies E Hulscher
2018 BMJ Open  
Downloaded from Berrevoets MAH, et al. BMJ Open 2018;8:e024564. doi:10.1136/bmjopen-2018-024564  ... 
doi:10.1136/bmjopen-2018-024564 pmid:30420352 fatcat:gdvzrwjrorai7mzhv34tjiadxi

The foundations of cost-sensitive causal classification [article]

Wouter Verbeke, Diego Olaya, Jeroen Berrevoets, Sam Verboven, Sebastián Maldonado
2021 arXiv   pre-print
Classification is a well-studied machine learning task which concerns the assignment of instances to a set of outcomes. Classification models support the optimization of managerial decision-making across a variety of operational business processes. For instance, customer churn prediction models are adopted to increase the efficiency of retention campaigns by optimizing the selection of customers that are to be targeted. Cost-sensitive and causal classification methods have independently been
more » ... posed to improve the performance of classification models. The former considers the benefits and costs of correct and incorrect classifications, such as the benefit of a retained customer, whereas the latter estimates the causal effect of an action, such as a retention campaign, on the outcome of interest. This study integrates cost-sensitive and causal classification by elaborating a unifying evaluation framework. The framework encompasses a range of existing and novel performance measures for evaluating both causal and conventional classification models in a cost-sensitive as well as a cost-insensitive manner. We proof that conventional classification is a specific case of causal classification in terms of a range of performance measures when the number of actions is equal to one. The framework is shown to instantiate to application-specific cost-sensitive performance measures that have been recently proposed for evaluating customer retention and response uplift models, and allows to maximize profitability when adopting a causal classification model for optimizing decision-making. The proposed framework paves the way toward the development of cost-sensitive causal learning methods and opens a range of opportunities for improving data-driven business decision-making.
arXiv:2007.12582v5 fatcat:vqr7c2libfgplov7gn2gkr6v4m

Relevance of neuroimaging for neurocognitive and behavioral outcome after pediatric traumatic brain injury

Marsh Königs, Petra JW Pouwels, LW Ernest van Heurn, Roel Bakx, R Jeroen Vermeulen, J Carel Goslings, Bwee Tien Poll-The, Marleen van der Wees, Coriene E Catsman-Berrevoets, Jaap Oosterlaan
2017 Brain Imaging and Behavior  
This study aims to (1) investigate the neuropathology of mild to severe pediatric TBI and (2) elucidate the predictive value of conventional and innovative neuroimaging for functional outcome. Children aged 8-14 years with trauma control (TC) injury (n = 27) were compared to children with mild TBI and risk factors for complicated TBI (mild RF+ , n = 20) or moderate/severe TBI (n = 17) at 2.8 years post-injury. Neuroimaging measures included: acute computed tomography (CT), volumetric analysis
more » ... post-acute conventional T1weighted magnetic resonance imaging (MRI) and post-acute diffusion tensor imaging (DTI, analyzed using tract-based spatial statistics and voxel-wise regression). Functional outcome was measured using Common Data Elements for neurocognitive and behavioral functioning. The results show that intracranial pathology on acute CT-scans was more prevalent after moderate/severe TBI (65%) than after mild RF+ TBI
doi:10.1007/s11682-017-9673-3 pmid:28092022 pmcid:PMC5814510 fatcat:yrxon65ffzdexn2zw4grfquefu

Impaired Visual Integration in Children with Traumatic Brain Injury: An Observational Study

Marsh Königs, Wouter D. Weeda, L. W. Ernest van Heurn, R. Jeroen Vermeulen, J. Carel Goslings, Jan S. K. Luitse, Bwee Tien Poll-Thé, Anita Beelen, Marleen van der Wees, Rachèl J. J. K. Kemps, Coriene E. Catsman-Berrevoets, Jaap Oosterlaan (+1 others)
2015 PLoS ONE  
Axonal injury after traumatic brain injury (TBI) may cause impaired sensory integration. We aim to determine the effects of childhood TBI on visual integration in relation to general neurocognitive functioning. Methods We compared children aged 6-13 diagnosed with TBI (n = 103; M = 1.7 years post-injury) to children with traumatic control (TC) injury (n = 44). Three TBI severity groups were distinguished: mild TBI without risk factors for complicated TBI (mild RF-TBI, n = 22), mild TBI with 1
more » ... sk factor (mild RF+ TBI, n = 46) or moderate/severe TBI (n = 35). An experimental paradigm measured speed and accuracy of goal-directed behavior depending on: (1) visual identification; (2) visual localization; or (3) both, measuring visual integration. Group-differences on reaction time (RT) or accuracy were tracked down to task strategy, visual processing efficiency and extra-decisional processes (e.g. response execution) using diffusion PLOS ONE | Data Availability Statement: All data on the group level is available within the paper and supporting information. On authority of the Data Protection Officer of the VU University Amsterdam, the medical data on the level of the individual subject is not made available: even the anonymized minimum dataset contains small groups that carry the risk of person identification and is therefore restricted from sharing under the Dutch Personal Data Protection Act. Therefore, the minimum dataset cannot be published online and also cannot be made available upon request. model analysis. General neurocognitive functioning was measured by a Wechsler Intelligence Scale short form. Results The TBI group had poorer accuracy of visual identification and visual integration than the TC group (Ps .03; ds -0.40). Analyses differentiating TBI severity revealed that visual identification accuracy was impaired in the moderate/severe TBI group (P = .05, d = -0.50) and that visual integration accuracy was impaired in the mild RF+ TBI group and moderate/ severe TBI group (Ps < .02, ds -0.56). Diffusion model analyses tracked impaired visual integration accuracy down to lower visual integration efficiency in the mild RF+ TBI group and moderate/severe TBI group (Ps < .001, ds -0.73). Importantly, intelligence impairments observed in the TBI group (P = .009, d = -0.48) were statistically explained by visual integration efficiency (P = .002).
doi:10.1371/journal.pone.0144395 pmid:26637182 pmcid:PMC4670090 fatcat:5ttyqbkk2rcija7thjrjwc6z4q

Monitoring, documenting and reporting the quality of antibiotic use in the Netherlands: a pilot study to establish a national antimicrobial stewardship registry

Marvin AH Berrevoets, Jaap ten Oever, Tom Sprong, Reinier M van Hest, Ingeborg Groothuis, Inger van Heijl, Jeroen A Schouten, Marlies E Hulscher, Bart-Jan Kullberg
2017 BMC Infectious Diseases  
The Dutch Working Party on Antibiotic Policy is developing a national antimicrobial stewardship registry. This registry will report both the quality of antibiotic use in hospitals in the Netherlands and the stewardship activities employed. It is currently unclear which aspects of the quality of antibiotic use are monitored by antimicrobial stewardship teams (A-teams) and can be used as indicators for the stewardship registry. In this pilot study we aimed to determine which stewardship
more » ... are eligible for the envisioned registry. Methods: We performed an observational pilot study among five Dutch hospitals. We assessed which of the 14 validated stewardship objectives (11 process of care recommendations and 3 structure of care recommendations) the A-teams monitored and documented in individual patients. They provided, where possible, data to compute quality indicator (QI) performance scores in line with recently developed QIs to measure appropriate antibiotic use in hospitalized adults for the period of January 2015 through December 2015 Results: All hospitals had a local antibiotic guideline describing recommended antimicrobial use. All A-teams monitored the performance of bedside consultations in Staphylococcus aureus bacteremia and the prescription of restricted antimicrobials. Documentation and reporting were the best for the use of restricted antimicrobials: 80% of the A-teams could report data. Lack of time and the absence of an electronic medical record system enabling documentation during the daily work flow were the main barriers hindering documentation and reporting. Conclusions: Five out of 11 stewardship objectives were actively monitored by A-teams. Without extra effort, 4 A-teams could report on the quality of use of restricted antibiotics. Therefore, this aspect of antibiotic use should be the starting point of the national antimicrobial stewardship registry. Our registry is expected to become a powerful tool to evaluate progress and impact of antimicrobial stewardship programs in hospitals.
doi:10.1186/s12879-017-2673-5 pmid:28806902 pmcid:PMC5557571 fatcat:t7b2feecuzgebmx3wbmr2kpr5u
« Previous Showing results 1 — 15 out of 23 results