Black-box Models’ Explainability: A Theoretical and Practical Perspective

2023 | Americas Conference on Information Systems | Citations: 0

Authors: He, Yuanqi; Gu, Bin; Li, Chunxiao; Yan, Lingling; Ding, Yanqing

Abstract: The lack of explainability remains a critical challenge to the widespread adopt ... Expand

Abstract: The lack of explainability remains a critical challenge to the widespread adoption of artificial intelligence (AI) in many fields. "Understanding brings in trust", while machine-learning models offer superior prediction accuracy, understanding the underlying logic is equally important to foster trust in these models. In this paper, we present eXplainable AI (XAI) as a solution to this challenge. Our research focuses on three key aspects of XAI: mathematics, humanities and social sciences, and practical applications. We demonstrated the feasibility of XAI through the use of artificially-constructed and model-derived ground truth, and verified performances of different XAIs. We also explored three dimensions of explainable consistency and emphasized the significance of human-machine consistency. Finally, we applied our research to a real-world scenario by cooperating with a national bank in China. Our findings highlight that XAI is both mathematically and practically meaningful, but more efforts need to be dedicated to this human-machine communication field. Collapse

Topics: explainable artificial intelligence shapley additive explanation local interpretable model agnostic explanation decision making accounting

Methods: computational algorithm principal component analysis data transformation XGBoost experiment

ROLEX: A Novel Method for Interpretable Machine Learning Using Robust Local Explanations

2023 | Management Information Systems Quarterly | Citations: 2

Authors: Kim, Buomsoo (Raymond); Srinivasan, Karthik; Kong, Sung Hye; Kim, Jung Hee; Shin, Chan Soo; Ram, Sudha

Abstract: Recent developments in big data technologies are revolutionizing the field of h ... Expand

Abstract: Recent developments in big data technologies are revolutionizing the field of healthcare predictive analytics (HPA), enabling researchers to explore challenging problems using complex prediction models. Nevertheless, healthcare practitioners are reluctant to adopt those models as they are less transparent and accountable due to their black-box structure. We believe that instance-level, or local, explanations enhance patient safety and foster trust by enabling patient-level interpretations and medical knowledge discovery. Therefore, we propose the RObust Local EXplanations (ROLEX) method to develop robust, instance-level explanations for HPA models in this study. ROLEX adapts state-of-the-art methods and ameliorates their shortcomings in explaining individual-level predictions made by black-box machine learning models. Our analysis with a large real-world dataset related to a prevalent medical condition called fragility fracture and two publicly available healthcare datasets reveals that ROLEX outperforms widely accepted benchmark methods in terms of local faithfulness of explanations. In addition, ROLEX is more robust since it does not rely on extensive hyperparameter tuning or heuristic algorithms. Explanations generated by ROLEX, along with the prototype user interface presented in this study, have the potential to promote personalized care and precision medicine by providing patient-level interpretations and novel insights. We discuss the theoretical implications of our study in healthcare, big data, and design science. Collapse

Topics: explainable artificial intelligence usability user interface local interpretable model agnostic explanation problem solving

Methods: machine learning design artifact computational algorithm longitudinal research ablation analysis

Theories: lemon market theory

A holistic approach to interpretability in financial lending: Models, visualizations, and summary-explanations

2022 | Decision Support Systems | Citations: 0

Authors: Chen, Chaofan; Lin, Kangcheng; Rudin, Cynthia; Shaposhnik, Yaron; Wang, Sijia; Wang, Tong

Abstract: A B S T R A C TLending decisions are usually made with proprietary models that ... Expand

Abstract: A B S T R A C TLending decisions are usually made with proprietary models that provide minimally acceptable explanations to users. In a future world without such secrecy, what decision support tools would one want to use for justified lending decisions? This question is timely, since the economy has dramatically shifted due to a pandemic, and a massive number of new loans will be necessary in the short term. We propose a framework for such decisions, including a globally interpretable machine learning model, an interactive visualization of it, and several types of summaries and explanations for any given decision. The machine learning model is a two-layer additive risk model, which resembles a two-layer neural network, but is decomposable into subscales. In this model, each node in the first (hidden) layer represents a meaningful subscale model, and all of the nonlinearities are transparent. Our online visualization tool allows exploration of this model, showing precisely how it came to its conclusion. We provide three types of explanations that are simpler than, but consistent with, the global model: case-based reasoning explanations that use neighboring past cases, a set of features that were the most important for the model's prediction, and summary-explanations that provide a customized sparse explanation for any particular lending decision made by the model. Our framework earned the FICO recognition award for the Explainable Machine Learning Challenge, which was the first public challenge in the domain of explainable machine learning. 1 Collapse

Topics: logistics management decision support system explainable artificial intelligence database system missing data

Methods: machine learning artificial neural network computational algorithm optimization model paired t-test

Theories: theory of case-based reasoning

Does a CDO enhance Company Inventiveness? An Empirical Study on new Managerial Positions and Innovation

2022 | Americas Conference on Information Systems | Citations: 0

Authors: R Salas, Ruben; Fernandez-Corrales, Carla

Abstract: Digital technology has advanced rapidly in the last few years. Consequently, co ... Expand

Abstract: Digital technology has advanced rapidly in the last few years. Consequently, companies need to transform their business and operations models to keep up with these advances, placing digital transformation (DT) as a top priority. Beyond the possible effects of using more digital technologies, is more innovation. One way in which companies can stimulate DT is by creating a new top-management position, the Chief Digital Officer (CDO). Although evidence has been found about the relationship between digital technologies and innovation, the role of the CDO as an enabler has been less studied. We use difference-in-difference estimators to measure this effect, using a unique dataset with 39 companies that appointed a CDO and 39 comparable companies that did not as a control group. We find a significant increase in patent filings as early as the first year CDO's appointment. We find evidence that a managerial role can have an effect on innovation. Collapse

Topics: innovation management digital transformation Chief Information Officer organizational change database system

Methods: difference in differences propensity score method experimental group longitudinal research linear regression analysis

Theories: technology acceptance theory

Evaluating the Effectiveness of Marketing Campaigns for Malls Using a Novel Interpretable Machine Learning Model

2022 | Information Systems Research | Citations: 1

Authors: Wang, Tong; He, Cheng; Jin, Fujie; Hu, Yu Jeffrey

Abstract: In this study, we use newly available data and develop a novel interpretable mac ... Expand

Abstract: In this study, we use newly available data and develop a novel interpretable machine learning model to evaluate how different types of marketing campaigns and budget allocations influence malls’ customer traffic. The data we use is a large-scale customer traffic data set, collected through AI-chip-embedded sensors, across 25 malls over a two-year period, and we combine it with detailed campaign information for our analyses. We classify the campaigns into five categories based on the approach and timing of the campaigns. We then develop an innovative interpretable machine learning model, named generalized additive neural network model (GANNM), to accurately learn the response curves for different marketing campaigns. The response curves characterize the impact of campaign budget on customer traffic. We demonstrate that this new model has better predictive accuracy compared with current interpretable models and also yields additional business insights. We find that campaigns with experience incentives lead to larger increases in customer traffic than campaigns with sales incentives only, and the contrast is more significant for campaigns in off-peak periods. In addition, malls can piggyback on online promotion events and boost customer traffic with campaigns held at the same time. We further demonstrate that the optimized budget allocation based on the response curves learned by GANNM yields a 11.2% increase in customer traffic overall, compared with 3.2% achieved by a baseline with preassumed functional forms of response curves and 1.0% achieved by a post hoc explanation method. Overall, our proposed model provides more accurate estimations for response curves and presents interpretable and actionable insights for managers. Collapse

Topics: marketing management explainable artificial intelligence artificial intelligence digital transformation innovation management

Methods: artificial neural network machine learning computational algorithm generalized additive model optimization model

GAM(e) changer or not? An evaluation of interpretable machine learning models based on additive model constraints

2022 | European Conference On Information Systems | Citations: 3

Authors: Zschech, Patrick; Weinzierl, Sven; Hambauer, Nico; Zilker, Sandra; Kraus, Mathias

Abstract: Mathias, "GAM(e) changer or not? An evaluation of interpretable machine learnin ... Expand

Abstract: Mathias, "GAM(e) changer or not? An evaluation of interpretable machine learning models based on additive model constraints" (2022). ECIS AbstractThe number of information systems (IS) studies dealing with explainable artificial intelligence (XAI) is currently exploding as the field demands more transparency about the internal decision logic of machine learning (ML) models. However, most techniques subsumed under XAI provide posthocanalytical ex planations, which have to be considered with caution as they only use approximations of the underlying ML model. Therefore, our paper investigates a series of intrinsically interpretable ML models and dis cusses their suitability for the IS community. More specifically, our focus is on advanced extensions of generalized additive models (GAM) in which predictors are modeled independently in a nonlinear way to generate shape functions that can capture arbitrary patterns but remain fully interpretable. In our study, we evaluate the prediction qualities of five GAMs as compared to six traditional ML models and assess their visual outputs for model interpretability. On this basis, we investigate their merits and limitations and derive design implications for further improvements. Collapse

Topics: Python explainable artificial intelligence missing data usability IS discipline

Methods: experiment design principle generalized additive model descriptive statistic archival research

Mis-spending on information security measures: Theory and experimental evidence

2021 | International Journal of Information Management | Citations: 0

Authors: Safi, Roozmehr; Browne, Glenn J.; Jalali Naini, Azadeh

Abstract: Information resources are becoming increasingly important to individuals and org ... Expand

Abstract: Information resources are becoming increasingly important to individuals and organizations, and ensuring their security is a major concern. While research in information security has adopted primarily a quantitative method to determine how and how much to invest in security, most decision makers rely on non-quantitative methods for this purpose, thereby introducing a considerable amount of as yet unexplained subjective judgment to the problem. We use a behavioral decision making approach to investigate factors causing possible inefficiencies of security spending decisions. Decision makers in our experiment performed a series of economic games featuring the key characteristics of a typical security problem. We found several biases in investment decisions. For budgeting their investment between major classes of security measures, decision makers demonstrated a strong bias toward investing in preventive measures rather than in detection and response measures, even though the task was designed to yield the same return on investment for both classes of measures. We term this phenomenon the “Prevention Bias.” Decision makers also reacted to security threats when the risk was so small that no in­ vestment was economically justified. For higher levels of risk that warranted some security investment, decision makers showed a strong tendency to overinvest. Theoretical and practical implications of the findings are discussed. Collapse

Topics: IT security defense decision making data security accounting IT security threat

Methods: experiment Student's t-test theory development parametric test statistical hypothesis test

Theories: loss aversion prospect theory behavioral economics

Finding a Needle in the Haystack: Recommending Online Communities on Social Media Platforms Using Network and Design Science

2021 | Journal of the Association for Information Systems | Citations: 0

Authors: Velichety, Srikar; Ram, Sudha

Abstract: We address the problem of recommending online communities on social media platf ... Expand

Abstract: We address the problem of recommending online communities on social media platforms using design science. Our method is grounded in network science and leverages the random surfer model of the web, small-world networks, strength of weak connections, and connectivity to analyze three types of large-scale networks. In doing so, we design features for structural hole assortativity and local clustering coefficient rank to capture both the diversity and evolution of user interests. We also extract general online community features such as size and overlap. Experiments conducted on a large dataset of 34,000 lists created and subscribed to by 1,600 active Twitter users over a six-month period showed that our network features outperform the general and content features in terms of recommending communities at the top position. In addition, a combination of general and network features generated the best results in the top position with a significant performance improvement over using only the content features. A combination of all three types of features gave the best results in the top-5 and top-10 positions while improving the quality of recommendations at every other position. Our work outperforms the latest work on community recommendations on social media platforms and has major implications for the design of online community recommenders. Collapse

Topics: Twitter social media online community recommender system big data

Methods: computational algorithm design science cluster analysis artificial neural network deep learning

A Bayesian Optimal Stopping Framework for Traumatic Brain Injuries Patients

2021 | International Conference on Information Systems | Citations: 0

Authors: Zavadskiy, Gleb; Zantedeschi, Daniel; Jank, Wolfgang

Abstract: One of the central veins of the Information System (IS) discipline is to help p ... Expand

Abstract: One of the central veins of the Information System (IS) discipline is to help practitioners in other fields to solve their real-world problems. This article focuses on building the decision support system for patients with Traumatic Brain Injuries (TBI) using Hidden Markov Models (HMM). Most of the existing literature focuses on the long-term outcomes arising from TBIs. Our development's key distinguishing factor is that we predict the likelihood of stability within the next 24 hours to undergo a non-cranial surgery of polytrauma patients with TBI in acute settings. Our goal in this work is to build an HMMbased Bayesian optimization framework to help doctors to determine the optimal release schedule for TBI patients. Collapse

Topics: decision making decision support system health information system

Methods: hidden Markov model Markov chain Monte Carlo logistic regression generalized additive model descriptive statistic

A Comprehensive Analysis of Triggers and Risk Factors for Asthma Based on Machine Learning and Large Heterogeneous Data Sources

2020 | Management Information Systems Quarterly | Citations: 0

Authors: Wenli Zhang; Ram, Sudha

Abstract: Asthma is a common chronic health condition affecting millions of people in the ... Expand

Abstract: Asthma is a common chronic health condition affecting millions of people in the United States. While asthma cannot be cured, it can be managed if we identify and understand triggers and risk factors that cause asthma exacerbations. However, this is challenging because these triggers and risk factors are complex and interconnected, and there are limitations to current mainstream approaches for identifying them. The recent availability of massive amounts of heterogeneous data has opened up new possibilities for asthma triggers and risk factors analyses. In this study, we introduce a data-driven framework, adapt and integrate multiple advanced machine learning techniques, and perform an empirical analysis to (1) derive characteristics of selfreported asthma patients from social media, (2) enable integration and repurposing of highly heterogeneous and commonly available datasets, and (3) uncover the sequential patterns of asthma triggers and risk factors, and their relative importance, both of which are difficult to achieve via retrospective cohort-based studies. Our methods and results can provide guidance for developing asthma management plans and interventions for specific subpopulations and, eventually, have the potential to reduce the societal burden of asthma. Collapse

Topics: social media Twitter database system open source wiki

Methods: random forest classification machine learning computational algorithm longitudinal research survey