Black-box Problems Research Articles

Multiple paths to enhancing the resilience of project-based organizations from the perspective of CSR configuration: evidence from the Chinese construction industry

PurposeThe role of corporate social responsibility (CSR) fulfillment is critical when building resilience of project-based organizations (PBOs). However, fulfilling CSR to build a highly resilient PBO remains a black box problem. This study explores the different CSR combinations that enhance PBO resilience.Design/methodology/approachThis study defines CSR in terms of shareholder, employee, and social CSR, and analyzes corporate characteristics in terms of corporate scale and nature. Data are collected from Hexun.com and the China Stock Market and Accounting Research Database (CSMAR). The qualitative comparative analysis (QCA) method is used to analyze 48 listed construction and engineering companies from China to explore the CSR configurations for PBOs in enhancing organizational resilience.FindingsA large firm size is a necessary condition for high organizational resilience. We find six paths to build high and non-high resilience in PBOs, and the driving mechanisms of high and non-high resilience exhibit an asymmetric relationship.Research limitations/implicationsThis study cracks the black box of CSR fulfillment and PBO resilience. It reveals the CSR configurations that enhance or inhibit the resilience of PBOs. It also provides scientific basis for PBOs in their fulfillment of CSR in response to crises, and the enhancement of organizational resilience. Future research can be expanded to other industries, as the study sample is only limited to civil engineering construction companies. Since this study uses cross-sectional data, time series can be introduced in the future to further explore the relationship between CSR and organizational resilience.Practical implicationsThis study provides targeted suggestions that can help decision-makers of construction companies to determine how they can fulfill CSR to enhance organizational resilience. At the same time, it can provide intellectual support for PBOs to cope with systemic crises and promote the fulfillment of CSR.Originality/valueIn terms of theoretical value, on the one hand, this study verifies the relationship between CSR fulfillment and PBO resilience, revealing its mechanism of action and multiple paths; on the other hand, it provides a new way of thinking for management research methods and enriches the theoretical study of organizational resilience.

TREGO: a trust-region framework for efficient global optimization

Efficient global optimization (EGO) is the canonical form of Bayesian optimization that has been successfully applied to solve global optimization of expensive-to-evaluate black-box problems. However, EGO struggles to scale with dimension, and offers limited theoretical guarantees. In this work, a trust-region framework for EGO (TREGO) is proposed and analyzed. TREGO alternates between regular EGO steps and local steps within a trust region. By following a classical scheme for the trust region (based on a sufficient decrease condition), the proposed algorithm enjoys global convergence properties, while departing from EGO only for a subset of optimization steps. Using extensive numerical experiments based on the well-known COCO bound constrained problems, we first analyze the sensitivity of TREGO to its own parameters, then show that the resulting algorithm is consistently outperforming EGO and getting competitive with other state-of-the-art black-box optimization methods.

Explainability of artificial intelligence methods, applications and challenges: A comprehensive survey

The continuous advancement of Artificial Intelligence (AI) has been revolutionizing the strategy of decision-making in different life domains. Regardless of this achievement, AI algorithms have been built as Black-Boxes, that is as they hide their internal rationality and learning methodology from the human leaving many unanswered questions about how and why the AI decisions are made. The absence of explanation results in a sensible and ethical challenge. Explainable Artificial Intelligence (XAI) is an evolving subfield of AI that emphasizes developing a plethora of tools and techniques for unboxing the Black-Box AI solutions by generating human-comprehensible, insightful, and transparent explanations of AI decisions. This study begins by discussing the primary principles of XAI research, Black-Box problems, the targeted audience, and the related notion of explainability over the historical timeline of the XAI studies and accordingly establishes an innovative definition of explainability that addresses the earlier theoretical proposals. According to an extensive analysis of the literature, this study contributes to the body of knowledge by driving a fine-grained, multi-level, and multi-dimension taxonomy for insightful categorization of XAI studies with the main aim to shed light on the variations and commonalities of existing algorithms paving the way for extra methodological developments. Then, an experimental comparative analysis is presented for the explanation generated by common XAI algorithms applied to different categories of data to highlight their properties, advantages, and flaws. Followingly, this study discusses and categorizes the evaluation metrics for the XAI-generated explanation and the findings show that there is no common consensus on how an explanation must be expressed, and how its quality and dependability should be evaluated. The findings show that XAI can contribute to realizing responsible and trustworthy AI, however, the advantages of interpretability should be technically demonstrated, and complementary procedures and regulations are required to give actionable information that can empower decision-making in real-world applications. Finally, the tutorial is crowned by discussing the open research questions, challenges, and future directions that serve as a roadmap for the AI community to advance the research in XAI and to inspire specialists and practitioners to take the advantage of XAI in different disciplines.

Nonlinear Time Series Fuzzy Regression for Developing Explainable Consumer Preferences’ Models Based on Online Comments

In modeling of consumer preferences based on online comments, nonlinearity and fuzziness exist in the relationship between the product design attribute and the consumer preference. On the other hand, consumer preferences are not static and changing over the time. Previous studies have proposed few approaches to model the variational consumer preferences based on online comments. However, the obtained models have black box problems and are not easy to be understood by humans as explicit models cannot be shown, which give rise to the research area of explainable artificial intelligence. Therefore, it is necessary to develop understandable and accurate consumer preferences’ models. In this article, a nonlinear time series fuzzy regression method is proposed to model the variational consumer preference based on online comments, which can generate a fuzzy dynamic consumer preference model with interactive terms, second-order and/or higher order terms. The datasets are first extracted from online comments using the sentiment analysis. Then, the polynomial structure of the fuzzy dynamic consumer preference model is established by using multiobjective chaos optimization algorithm. Then, the fuzzy regression method is used to allocate the fuzzy coefficients of each item of the model. Using sweeping robot as a case study, the validation results from the proposed approach are compared with those from fuzzy least squares regression, time series fuzzy least squares regression, fuzzy regression, and time series fuzzy regression, and it is found that the proposed approach performs better than the other four approaches in terms of mean relative errors and mean system credibility.

A trust region based local Bayesian optimization without exhausted optimization of acquisition function

Bayesian optimization (BO) is an effective optimization technique for solving expensive black-box problems. Even though BO has remarkable success, its drawbacks are also obvious. First, the time complexity of the Gaussian process inference is higher than O(n3), where n is the number of samples. Consequently, the running time of BO increases rapidly with the problem size. Second, due to the non-convexity and multimodality of the acquisition function, it costs a lot to achieve good results. To address the above problems, we develop a local Bayesian optimization algorithm based on the trust region idea (TRLBO). In TRLBO, two trust regions with dynamically changing sizes are used to enhance the algorithm’s exploitation ability, while at the same time retaining the exploration ability. Specifically, one trust region is used to reduce the number of samples in the Gaussian process. The other is used to restrict the solution space of the candidates. Furthermore, some theoretical results were provided to enlighten the efficiency of the proposed algorithm. Experimental results on both benchmark functions and real-world problems show that TRLBO compares favorably with the state-of-the-art algorithms.

Multi-objective simulation–optimization via kriging surrogate models applied to natural gas liquefaction process design

A surrogate-based multi-objective optimization framework is employed in the design of natural gas liquefaction processes using reliable, black-box process simulation. The conflicting objectives are minimizing both power consumption and heat exchanger area utilization. The Pareto solutions of the single-mixed refrigerant (SMR) and propane-precooled mixed refrigerant (C3MR) processes are compared to determine the suitability of each process in terms of energy consumption and heat exchanger area. Kriging models and the ɛ-constraint methodology are used to sequentially provide simple surrogate optimization subproblems, whose minimizers are promising feasible and non-dominated solutions to the original black-box problem. The surrogate-based ɛ-constrained optimization subproblems are solved in GAMS using CONOPT. The Pareto Fronts achieved with the surrogate-based framework dominate the results from the NSGA-II, a well-established meta-heuristics of multi-objective optimization. The objective functions of non-dominated solutions go as low as 1045 and 980.3 kJ/kg-LNG and specific UA values of 212.2 and 266.9 kJ/(°C kg-LNG) for SMR and C3MR, respectively. The trade-off solutions that present the minimum sum of relative objectives are analyzed as well as the dominance of C3MR over SMR at low power consumption values and conversely at low heat exchanger area utilization.

Algorithm 1027: NOMAD Version 4: Nonlinear Optimization with the MADS Algorithm

NOMAD is a state-of-the-art software package for optimizing blackbox problems. In continuous development since 2001, it constantly evolved with the integration of new algorithmic features published in scientific publications. These features are motivated by real applications encountered by industrial partners. The latest major release of NOMAD , version 3, dates to 2008. Minor releases are produced as new features are incorporated. The present work describes NOMAD 4, a complete redesign of the previous version, with a new architecture providing more flexible code, added functionalities, and reusable code. We introduce algorithmic components, which are building blocks for more complex algorithms and can initiate other components, launch nested algorithms, or perform specialized tasks. They facilitate the implementation of new ideas, including the MegaSearchPoll component, warm and hot restarts, and a revised version of the PsdMads algorithm. Another main improvement of NOMAD 4 is the usage of parallelism, to simultaneously compute multiple blackbox evaluations and to maximize usage of available cores. Running different algorithms, tuning their parameters, and comparing their performance for optimization are simpler than before, while overall optimization performance is maintained between versions 3 and 4. NOMAD is freely available at www.gerad.ca/nomad and the whole project is visible at github.com/bbopt/nomad.

Co-Learning Bayesian Optimization.

Bayesian optimization (BO) is well known to be sample efficient for solving black-box problems. However, BO algorithms may get stuck in suboptimal solutions even with plenty of samples. Intrinsically, such a suboptimal problem of BO can attribute to the poor surrogate accuracy of the trained Gaussian process (GP), particularly that in the regions where the optimal solutions locate. Hence, we propose to build multiple GP models instead of a single GP surrogate to complement each other, thus resolving the suboptimal problem of BO. Nevertheless, according to the bias-variance tradeoff equation, the individual prediction errors can increase when increasing the diversity of models, which may lead even worse overall surrogate accuracy. On the other hand, based on the theory of the Rademacher complexity, it has been proven that exploiting the agreement of models on unlabeled information can reduce the complexity of hypothesis space, therefore achieving the required surrogate accuracy with fewer samples. Such value of model agreement has been extensively demonstrated for co-training style algorithms to boost model accuracy with a small portion of samples. Inspired by the above, we propose a novel BO algorithm labeled as co-learning BO (CLBO), which exploits both model diversity and agreement on unlabeled information to improve the overall surrogate accuracy with limited samples, therefore achieving more efficient global optimization. Through tests on five numerical toy problems and three engineering benchmarks, the effectiveness of the proposed CLBO has been well demonstrated.

Cooperative Bayesian optimization with hybrid grouping strategy and sample transfer for expensive large-scale black-box problems

Expensive large-scale optimization problems are difficult to solve owing to their high complexity and computational costs. Bayesian optimization is an effective method for dealing with low-dimensional, costly problems. This study extends Bayesian optimization to expensive large-scale optimization problems. In the study, the large-scale problem is decomposed into a series of low-dimensional subproblems using the proposed hybrid grouping strategy. Each subproblem is chosen based on its contribution to the improvement of the best overall objective value, and then Bayesian optimization is used to solve it. This study proposes a sample transfer strategy and a novel transfer Gaussian process regression model to improve the efficiency of Bayesian optimization. The sample transfer strategy can take advantage of the additive separability of the objective and thus transfers the historical samples by adding a constant to their objective values. The transfer Gaussian process regression model can employ the historical samples by reducing their correlations according to the differences in their parameters (i.e., variables other than those of the current subproblem). The proposed algorithm is tested on 15 large-scale benchmark problems up to 1000 dimensions. The experimental results show that the proposed algorithm outperforms the state-of-the-art algorithms, and the proposed hybrid grouping strategy, sample transfer strategy, and transfer Gaussian process regression improve the efficiency of the cooperative Bayesian optimization framework even further.

Bifidelity Surrogate Modelling: Showcasing the Need for New Test Instances

In recent years, multifidelity expensive black-box (Mf-EBB) methods have received increasing attention due to their strong applicability to industrial design problems. The challenge, however, is that knowledge of the relationship between decisions and objective values is limited to a small set of sample observations of variable quality. In the field of Mf-EBB, a problem instance consists of an expensive yet accurate source of information, and one or more cheap yet less accurate sources of information. The field aims to provide techniques either to accurately explain how decisions affect design outcome, or to find the best decisions to optimise design outcomes. Many techniques that use surrogate models have been developed to provide solutions to both aims. Only in recent years, however, have researchers begun to explore the conditions under which these new techniques are reliable, often focusing on problems with a single low-fidelity function, known as bifidelity expensive black-box (Bf-EBB) problems. This study extends the existing Bf-EBB test instances found in the literature, as well as the features used to determine when the low-fidelity information source should be used. A literature test suite is constructed and augmented with new instances to demonstrate the potentially misleading results that could be reached using only the instances currently found in the literature, and to expose the criticality of a more heterogeneous test suite for algorithm assessment. Addressing the shortcomings of the existing literature, a new set of features is presented, as well as a new instance creation procedure, and a study of their impact on algorithm assessment is conducted. The low-fidelity information source is shown to be valuable if it is often locally accurate, even when its overall accuracy is relatively low. This contradicts the existing literature guidelines, which indicate the low-fidelity information is only useful if it has a high overall accuracy. History: Accepted by Antonio Frangioni, Area Editor for Design & Analysis of Algorithms – Continuous. Funding: This work was supported by Australian Research Council [Grant IC200100009] for the ARC Training Centre in Optimisation Technologies, Integrated Methodologies and Applications (OPTIMA), and the University of Melbourne Research Computing Services and Petascale Campus Initiative. N. Andrés-Thió is also supported by a Research Training Program scholarship from the University of Melbourne. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplementary Information [ https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2022.1217 ] or is available from the IJOC GitHub software repository ( https://github.com/INFORMSJoC ) at [ http://dx.doi.org/10.5281/zenodo.6578060 ].

A Historical Experience Surrogate Model Assisted Particle Swarm Optimization for Expensive Black-box Problems

This paper proposed a new historical experience surrogate model assisted particle swarm optimization method. This method extends the particle swarm optimization by adding new surrogate-based phase. In the classic phase, the particle swarm optimization runs the same way as the original algorithm, and the real function value evaluated are collected into the global database. In the surrogate phase, sub-swarms are generated following the distribution of the history data and evaluated by the surrogate model(s). The purpose of the surrogate phase is to explore the possible better solutions of the searching history. Also, the surrogate model(s) have the ability of accelerating the intelligence algorithms. Nevertheless, considering the time complexity of training and evaluating the surrogate model(s), the original problem should be expensive to evaluate or driven by data, which are same as many real-world problems.

An efficient global optimization algorithm for expensive constrained black-box problems by reducing candidate infilling region

Surrogate-based optimization methods are usually employed to solve expensive black-box optimization problems. Constraints bring more challenges for optimization than unconstrained problems. This study proposes a novel three-stage infilling framework to solve expensive single-objective problems with equality and/or inequality constraints. In the first stage, a criterion for locating a feasible region is designed. Until a feasible solution is found, the first stage is removed in the subsequent process. The second stage exploits the current best feasible solution and explores potentially better solutions. If the infilled site is infeasible, the third stage will proceed to improve the surrogate accuracy around the feasible region boundary and an extra constraint derived from the objective is proposed to reduce the candidate infill region in this stage. Otherwise, the process stays in stage two. Stage two continues once one third-stage point has been sampled. For brevity, the three-stage framework is called CEGO-DI. The efficacy of CEGO-DI is demonstrated by comparing it with two variants of it. In addition, the method is also compared with several state-of-the-art algorithms on extensive problems. The experimental results show that CEGO-DI has a better or competitive performance on most problems.

Automatic False Alarm Detection Based on XAI and Reliability Analysis

Many studies attempt to apply artificial intelligence (AI) to cyber security to effectively cope with the increasing number of cyber threats. However, there is a black box problem such that it is difficult to understand the basis for AI prediction. False alarms for malware or cyberattacks can cause serious side effects. Due to this limitation, all AI predictions must be confirmed by an expert, which is a considerable obstacle to AI expansion. Compared to the increasing number of cyberattack alerts, the number of alerts that can be analyzed by experts is limited. This paper provides explainability through an interpretation of AI prediction results and a reliability analysis of AI predictions based on explainable artificial intelligence (XAI). In addition, we propose a method for screening high-quality data that can efficiently detect false predictions based on reliability indicators. Through this, even a small security team can quickly respond to false predictions. To validate the proposed method, experiments were conducted using the IDS dataset and the malware dataset. AI errors were detected better than they could be by the existing AI models, with about 262% in the IDS dataset and 127% in the malware dataset from the top 10% of analysis targets. Therefore, the ability to respond to cyberattacks can be improved using the proposed method.

Diagnosis of Paratuberculosis in Histopathological Images Based on Explainable Artificial Intelligence and Deep Learning

Artificial intelligence holds great promise in medical imaging, especially histopathological imaging. However, artificial intelligence algorithms cannot fully explain the thought processes during decision-making. This situation has brought the problem of explainability, i.e., the black box problem, of artificial intelligence applications to the agenda: an algorithm simply responds without stating the reasons for the given images. To overcome the problem and improve the explainability, explainable artificial intelligence (XAI) has come to the fore, and piqued the interest of many researchers. Against this backdrop, this study examines a new and original dataset using the deep learning algorithm, and visualizes the output with gradient-weighted class activation mapping (Grad-CAM), one of the XAI applications. Afterwards, a detailed questionnaire survey was conducted with the pathologists on these images. Both the decision-making processes and the explanations were verified, and the accuracy of the output was tested. The research results greatly help pathologists in the diagnosis of paratuberculosis.

Pathway importance by graph convolutional network and Shapley additive explanations in gene expression phenotype of diffuse large B-cell lymphoma.

Deep learning techniques have recently been applied to analyze associations between gene expression data and disease phenotypes. However, there are concerns regarding the black box problem: it is difficult to interpret why the prediction results are obtained using deep learning models from model parameters. New methods have been proposed for interpreting deep learning model predictions but have not been applied to genetics. In this study, we demonstrated that applying SHapley Additive exPlanations (SHAP) to a deep learning model using graph convolutions of genetic pathways can provide pathway-level feature importance for classification prediction of diffuse large B-cell lymphoma (DLBCL) gene expression subtypes. Using Kyoto Encyclopedia of Genes and Genomes pathways, a graph convolutional network (GCN) model was implemented to construct graphs with nodes and edges. DLBCL datasets, including microarray gene expression data and clinical information on subtypes (germinal center B-cell-like type and activated B-cell-like type), were retrieved from the Gene Expression Omnibus to evaluate the model. The GCN model showed an accuracy of 0.914, precision of 0.948, recall of 0.868, and F1 score of 0.906 in analysis of the classification performance for the test datasets. The pathways with high feature importance by SHAP included highly enriched pathways in the gene set enrichment analysis. Moreover, a logistic regression model with explanatory variables of genes in pathways with high feature importance showed good performance in predicting DLBCL subtypes. In conclusion, our GCN model for classifying DLBCL subtypes is useful for interpreting important regulatory pathways that contribute to the prediction.

Expectile regression forest: A new nonparametric expectile regression model

AbstractClassical nonlinear expectile regression has two shortcomings. It is difficult to choose a nonlinear function, and it does not consider the interaction effects among explanatory variables. Therefore, we combine the random forest model with the expectile regression method to propose a new nonparametric expectile regression model: expectile regression forest (ERF). The major novelty of the ERF model is using the bagging method to build multiple decision trees, calculating the conditional expectile of each leaf node in each decision tree, and deriving final results through aggregating these decision tree results via simple average approach. At the same time, in order to compensate for the black box problem in the model interpretation of the ERF model, the measurement of the importance of explanatory variable and the partial dependence is defined to evaluate the magnitude and direction of the influence of each explanatory variable on the response variable. The advantage of ERF model is illustrated by Monte Carlo simulation studies. The numerical simulation results show that the estimation and prediction ability of the ERF model is significantly better than alternative approaches. We also apply the ERF model to analyse the real data. From the nonparametric expectile regression analysis of these data sets, we have several conclusions that are consistent with the results of numerical simulation.

Karl Jaspers and artificial neural nets: on the relation of explaining and understanding artificial intelligence in medicine

Assistive systems based on Artificial Intelligence (AI) are bound to reshape decision-making in all areas of society. One of the most intricate challenges arising from their implementation in high-stakes environments such as medicine concerns their frequently unsatisfying levels of explainability, especially in the guise of the so-called black-box problem: highly successful models based on deep learning seem to be inherently opaque, resisting comprehensive explanations. This may explain why some scholars claim that research should focus on rendering AI systems understandable, rather than explainable. Yet, there is a grave lack of agreement concerning these terms in much of the literature on AI. We argue that the seminal distinction made by the philosopher and physician Karl Jaspers between different types of explaining and understanding in psychopathology can be used to promote greater conceptual clarity in the context of Machine Learning (ML). Following Jaspers, we claim that explaining and understanding constitute multi-faceted epistemic approaches that should not be seen as mutually exclusive, but rather as complementary ones as in and of themselves they are necessarily limited. Drawing on the famous example of Watson for Oncology we highlight how Jaspers’ methodology translates to the case of medical AI. Classical considerations from the philosophy of psychiatry can therefore inform a debate at the centre of current AI ethics, which in turn may be crucial for a successful implementation of ethically and legally sound AI in medicine.

A Nested Weighted Tchebycheff Multi-Objective Bayesian Optimization Approach for Flexibility of Unknown Utopia Estimation in Expensive Black-Box Design Problems

Abstract We propose a nested weighted Tchebycheff Multi-objective Bayesian optimization (WTB MOBO) framework where we built a regression model selection procedure from the ensemble of models, toward better estimation of the uncertain parameters (utopia) of the weighted Tchebycheff expensive black-box multi-objective function. In our previous work, a weighted Tchebycheff MOBO approach has been demonstrated which attempts to estimate the model parameters (utopia) in formulating the acquisition function of the weighted Tchebycheff multi-objective black-box functions, through calibration using an a priori selected regression model. However, the existing MOBO model lacks flexibility in selecting the appropriate regression models given the guided sampled data and, therefore, can under-fit or over-fit as the iterations of the MOBO progress. This ultimately can reduce the overall MOBO performance. As, in general, it is too complex to a priori guarantee a best model, this motivates us to consider a portfolio of different families (simple-to-complex) of predictive models that have been fitted with current training data guided by the WTB MOBO, and the best model is selected following a user-defined prediction root-mean-square error-based approach. The proposed approach is implemented in optimizing a thin tube design under constant loading of temperature and pressure, minimizing the risk of creep-fatigue failure and design cost. Finally, the nested WTB MOBO model performance is compared with different MOBO frameworks with respect to accuracy in parameter estimation, Pareto-optimal solutions, and function evaluation cost. This approach is generalized enough to consider different families of predictive models in the portfolio for best model selection, where the overall design architecture allows for solving any high-dimensional (multiple functions) complex black-box problems and can be extended to any other global criterion multi-objective optimization methods where prior knowledge of utopia is required.

Xprediction: Explainable EGFR-TKIs response prediction based on drug sensitivity specific gene networks

In recent years, drug sensitivity prediction has garnered a great deal of attention due to the growing interest in precision medicine. Several computational methods have been developed for drug sensitivity prediction and the identification of related markers. However, most previous studies have ignored genetic interaction, although complex diseases (e.g., cancer) involve many genes intricately connected in a molecular network rather than the abnormality of a single gene. To effectively predict drug sensitivity and understand its mechanism, we propose a novel strategy for explainable drug sensitivity prediction based on sample-specific gene regulatory networks, designated Xprediction. Our strategy first estimates sample-specific gene regulatory networks that enable us to identify the molecular interplay underlying varying clinical characteristics of cell lines. We then, predict drug sensitivity based on the estimated sample-specific gene regulatory networks. The predictive models are based on machine learning approaches, i.e., random forest, kernel support vector machine, and deep neural network. Although the machine learning models provide remarkable results for prediction and classification, we cannot understand how the models reach their decisions. In other words, the methods suffer from the black box problem and thus, we cannot identify crucial molecular interactions that involve drug sensitivity-related mechanisms. To address this issue, we propose a method that describes the importance of each molecular interaction for the drug sensitivity prediction result. The proposed method enables us to identify crucial gene-gene interactions and thereby, interpret the prediction results based on the identified markers. To evaluate our strategy, we applied Xprediction to EGFR-TKIs prediction based on drug sensitivity specific gene regulatory networks and identified important molecular interactions for EGFR-TKIs prediction. Our strategy effectively performed drug sensitivity prediction compared with prediction based on the expression levels of genes. We also verified through literature, the EGFR-TKIs-related mechanisms of a majority of the identified markers. We expect our strategy to be a useful tool for predicting tasks and uncovering complex mechanisms related to pharmacological profiles, such as mechanisms of acquired drug resistance or sensitivity of cancer cells.

Machine learning interpretability for a stress scenario generation in credit scoring based on counterfactuals

To boost the application of machine learning (ML) techniques for credit scoring models, the blackbox problem should be addressed. The primary aim of this paper is to propose a measure based on counterfactuals to evaluate the interpretability of a ML credit scoring technique. Counterfactuals assist with understanding the model with regard to the classification decision boundaries and evaluate model robustness. The second contribution is the development of a data perturbation technique to generate a stress scenario.We apply these two proposals to a dataset on UK unsecured personal loans to compare logistic regression and stochastic gradient boosting (SBG). We show that training a blackbox model (SGB) as conditioned on our data perturbation technique can provide insight into model performance under stressed scenarios. The empirical results show that our interpretability measure is able to capture the classification decision boundary, unlike AUC and the classification accuracy widely used in the banking sector.