Search

An Artificial Intelligence algorithm trained on data that reflect racial biases may yield racially biased outputs, even if the algorithm on its own is unbiased. For example, algorithms used to schedule medical appointments in the USA predict that Black patients are at a higher risk of no-show than non-Black patients, though technically accurate given existing data that prediction results in Black patients being overwhelmingly scheduled in appointment slots that cause longer wait times than non-Black patients. This perpetuates racial inequity, in this case lesser access to medical care. This gives rise to one type of Accuracy-Fairness trade-off: preserve the efficiency offered by using AI to schedule appointments or discard that efficiency in order to avoid perpetuating ethno-racial disparities. Similar trade-offs arise in a range of AI applications including others in medicine, as well as in education, judicial systems, and public security, among others. This article presents a framework for addressing such trade-offs where Machine Learning and Optimization components of the algorithm are decoupled. Applied to medical appointment scheduling, our framework articulates four approaches intervening in different ways on different components of the algorithm. Each yields specific results, in one case preserving accuracy comparable to the current state-of-the-art while eliminating the disparity.

Members of an online community peer-produce digital artifacts by negotiating different perspectives and personal knowledge bases. These negotiations are manifested in the temporal evolution of the peer-produced artifact. In this study, we conceptualize the evolution of a digital artifact as a trajectory in a feature space. Our theoretical frame suggests that, through negotiations, contributors’ actions “pull” the trajectory and shape its movement in the feature space. We hypothesize that the type of contributors that work on a focal article influences the extent to which that article’s trajectory explores alternative positions within that space, and that the trajectory’s exploration is, in turn, associated with the artifact’s quality. To test these hypotheses, we analyzed the trajectories of wiki articles drawn from two peer-production communities, Wikipedia and Wikia, tracking the evolution of 242 paired articles for over a decade during which the articles went through 536,745 revisions. We found that the contributors who are the most likely to increase the trajectory’s exploration are those that (1) return to work on the focal artifact and (2) are unregistered members in the broader online community. Further, our results show that the trajectory’s exploration has a curvilinear association with article quality, indicating that exploration contributes positively to quality, but that the effect is reversed when exploration exceeds a certain level. The insights derived from this study highlight the value of an artifact-centric approach to increasing our understanding of the dynamics underlying peer-production.

The return of a product is often one of a series of transactions that a consumer undertakes in search of a good. In this article, the authors analyze returns as part of a product search process: Upon returning a product, consumers may immediately purchase an alternative one, which they may later replace with another product, and so on, until they either ultimately keep their last purchase (Keep outcome) or not (No-keep outcome). Such a sequence of transactions is called a “product return episode”. In this work, the authors study consumer Keep and return abuse behavior using episodic metrics. Using data from a consumer electronics retailer, the authors show that analysis of product returns with episodic metrics provides insights that differ from, and go beyond, analyses with commonly-used transactional metrics. They find that although higher average price and larger store assortment at a subcategory level both tend to increase the return probability, they also increase the probability of keeping a product at the end of an episode, which points to profit-improving opportunities for retailers by allowing returns and tracking episodes. They also find that episodic metrics are useful for identifying return abuse.

Electric vehicles have become a trend as a replacement to gasoline-powered vehicles, and been promoted by worldwide policy makers as a solution to combat environmental problems and stimulate economy, whereas the lack of extreme fast charging infrastructure has become one main obstacle to broad adoption of electric vehicles. To promote the commercial success of electric vehicles, millions of extreme fast charging stations are expected to be established in the next decade in the U.S. However, the widespread of electric vehicles and extreme fast charging infrastructure would impact transportation systems, such as mobility, congestion, equity, toll revenue, and infrastructure maintenance needs. As transportation infrastructure can last for a long time, effective planning must be able to predict the impact of projects and policies decades into the future. Given the nature of transportation systems that is of multiple interacting systems, this study develops mathematical models to quantitatively analyze this complicated and important problem. In particular, we utilize game theory to understand the dependency between the choices made by travelers, and the congestion and delay in the system with respect to traffic assignment. Our research results provide insights and guidance to strategic planning of extreme fast charging infrastructure.

Although data mining problems require a flat mining table as input, in many real-world applications analysts are interested in finding patterns in a relational database. To this end, new methods and software have been recently developed that automatically add attributes (or features) to a target table of a relational database which summarize information from all other tables. When attributes are automatically constructed by these methods, selecting the important attributes is particularly difficult, because a large number of the attributes are highly correlated. In this setting, attribute selection techniques such as the Least Absolute Shrinkage and Selection Operator (lasso), elastic net, and other machine learning methods tend to under-perform. In this paper, we introduce a novel attribute selection procedure, where after an initial screening step, we cluster the attributes into different groups and apply the group lasso to select both the true attributes groups and then the true attributes. The procedure is particularly suited to high dimensional data sets where the attributes are highly correlated. We test our procedure on several simulated data sets and a real-world data set from a marketing database. The results show that our proposed procedure obtains a higher predictive performance while selecting a much smaller set of attributes when compared to other state-of-the-art methods.

Problem definition: Machine learning is often employed in appointment scheduling to identify the patients with the greatest no-show risk, so as to schedule them into or right after overbooked slots. That scheduling decision maximizes the clinic performance, as measured by a weighted sum of all patients’ waiting time and the provider’s overtime and idle time. However, if a racial group is characterized by a higher no-show risk, then the patients belonging to that racial group will be scheduled into or right after overbooked slots disproportionately to the general population. Academic/Practical Relevance: That scheduling decision is problematic because patients scheduled in those slots tend to have a worse service experience than the other patients, as measured by the time they spend in the waiting room. Thus, the challenge becomes minimizing the schedule cost while avoiding racial disparities. Methodology: Motivated by the real-world case of a large specialty clinic whose black patients have a higher no-show probability than non-black patients, we analytically study racial disparity in this context. Then, we propose new objective functions that minimize both schedule cost and racial disparity and that can be readily adopted by researchers and practitioners. We develop a race-aware objective, which instead of minimizing the waiting times of all patients, minimizes the waiting times of the racial group expected to wait the longest. We also develop race-unaware methodologies that do not consider race explicitly. We validate our findings both on simulated and real-world data. Results: We demonstrate that state-of-the-art scheduling systems cause the black patients in our data set to wait about 30% longer than nonblack patients. Our race-aware methodology achieves both goals of eliminating racial disparity and obtaining a similar schedule cost as that obtained by the state-of-the-art scheduling method, whereas the race-unaware methodologies fail to obtain both efficiency and fairness. Managerial implications: Our work uncovers that the traditional objective of minimizing schedule cost may lead to unintended racial disparities. Both efficiency and fairness can be achieved by adopting a race-aware objective. History: This paper has been accepted for the Manufacturing & Service Operations Management Special Section on Responsible Research in Operations Management. Funding: M. Samorani’s work was partly funded by the 2020 Leavey Research Grant. Supplemental Material: The online supplement is available at https://doi.org/10.1287/msom.2021.0999 .

[enter Abstract Body]Appointment no-shows are disruptive to healthcare clinics, and may increase patient waiting time and clinic overtime, resulting in increased clinic costs. Appointment scheduling models typically mitigate the negative effects of no-shows through appointment overbooking. Recent work has proposed a predictive overbooking framework, where a probabilisitic classifier predicts the no-show probability of individual appointment requests, and a scheduling algorithm uses those predictions to optimally schedule appointments. Because predicting no-shows is typically an imbalanced classification problem, the preferred classifier is often chosen based upon the area under the receiver operator characteristic curve (AUC), which is a commonly used metric for many other imbalanced classification problems. Contrary to intuition, in this paper we show that employing the AUC to select a classifier results in significantly lower schedule efficiency than using other metrics such as Log Loss or Brier Score. Our computational experiments, validated on large real-world appointment data, suggest that by using Log Loss or Brier Score instead of AUC, practitioners can improve the schedule quality by up to 7%.

Cloud-based workflow architecture has been widely used in e-science, e-business, smart city, and others, to automate business processes and improve their flexibility and maintainability. Online workflow executes in a collaborative and distributed environment and is prone to fraud and information leakage. Workflow authorization models are implemented to ensure that tasks are performed by authorized subjects with compliance of security/privacy polices. However, existing workflow authorization models have some limitations. First, most of the existing research focuses on static workflows, where an order arriving at a workflow traverses tasks in a fixed sequence. In many real applications, however, task routing is not deterministic, having a probability distribution or pattern that may be estimated from historical data. Second, existing research ignores practical resource constraints, like user utilization, order waiting time, etc. To address the limitations, this article studies the workflow authorization model under the more realistic dynamic settings. We formulate a workflow as a queueing system, so business constraints can be analytically represented, under reasonable assumptions. We model the studied problems as pseudo-Boolean satisfiaiblity problems and investigate their theoretical properties. We also develop algorithms and carry out computational studies. The experimental results show the effectiveness and efficiency of our developed solutions. Our research results are useful for production and process design in many real-life settings such as health care, online banking and electronic payment systems.

Even though patients often arrive early and out of turn for scheduled appointments in outpatient clinics, no research has been undertaken to establish whether an available provider should see an early patient right away (preempt) or wait for the patient scheduled next. This problem, which we call the “Wait-Preempt Dilemma,” is particularly relevant for “high-service-level” clinics (such as psychotherapy, chiropractic, acupuncture), where preempting may cause the missing patient to wait for an excessively long time, should she show up soon. Typically, the dilemma is resolved by preemption, where the provider starts serving the patient who has already arrived to avoid staying idle. By contrast, we analytically determine the time intervals where it is optimal to preempt and those where it is optimal to wait, and find that in some cases the provider should in fact stay idle, even in the presence of waiting patients. Our results suggest that the proposed analytical method outperforms the always-preempt policy in clinics that do not overbook and have service times longer than 30 minutes. In these cases, the analytical method dramatically reduces patient waiting times at the cost of a modest increase in overtime. By contrast, in clinics that overbook or have short service times, the two policies perform similarly, and hence the always-preempt policy is preferable due to its simplicity. A software application is provided that clinics can readily use to solve the wait-preempt dilemma.

Although statistical and machine learning methods require the input data to be in a tabular format, in real-world applications data are often stored across several tables in a relational database. How to build a single mining table from a relational database is a critical pre-processing step of any classification method, because including the right attributes may dramatically boost the accuracy of the classifier. We propose a methodology and implement a software program, Dataconda, to automatically mine a relational database. The user selects a class attribute contained in a table of the database and the procedure builds and selects predictors by exploring the whole database and aggregating information, without any user intervention. For example, our procedure may find that the best predictor for “product return” is the proportion of products returned by the same customer in the past, even if the user has not built any such attribute. Our procedure produces more expressive attributes than existing methods. Our experiments on the ISMS Durable Goods Datasets, a publicly available data set of product returns in retailing, suggest that our method allows new knowledge to emerge.

Often, the return of a product is merely one of a series of transactions that a consumer undertakes in search of a good. Recognizing this, whereas the extant literature views each product return transaction in isolation, we analyze returns as part of a search process: Upon returning a product, customers may immediately purchase an alternative one, which they may later replace with another product, and so on, until they either ultimately keep their last purchase (Keep outcome) or not (No-keep outcome). We call such a sequence of transactions a product return episode. Using data from a consumer electronics retailer, we show that episodic metrics of return behavior better explain managerially relevant measures, such as net customer spending, and offer a richer understanding of returns than commonly used transactional metrics. Our findings suggest that although higher price and larger store assortment both tend to increase the return probability, they also increase the probability of keeping a product at the end of an episode, which may point to profit-improving opportunities for retailers.

Drug discovery is the process of designing compounds that have desirable properties, such as activity and nontoxicity. Molecule classification techniques are used along with this process to predict the properties of the compounds to expedite their testing. Ideally, the classification rules found should be accurate and reveal novel chemical properties, but current molecule representation techniques lead to less-than-adequate accuracy and knowledge discovery. This work extends the propositionalization approach recently proposed for multirelational data mining in two ways: it generates expressive attributes exhaustively, and it uses randomization to sample a limited set of complex (“deep”) attributes. Our experimental tests show that the procedure is able to generate meaningful and interpretable attributes from molecular structural data, and that these features are effective for classification purposes.

Even though it is well known that patients often arrive early and out of turn for scheduled appointments in outpatient clinics, no research has been undertaken to establish whether an idle provider should see an early patient right away (preempt) or wait for the patient scheduled next. In practice, this problem is typically resolved using a first-come-first-served policy, which consists of seeing the early patient. By contrast, we analytically determine the time intervals where it is optimal to preempt and those where it is optimal to wait. Our analysis indicates that the first-come-first-served policy is never optimal, although it is a good heuristic under certain circumstances, e.g. when overtime charges are high, appointment lengths are short, or if patients are likely to arrive very late. Compared to the first-come-first-served policy, our method dramatically reduces patient waiting times at the cost of a modest increase in overtime. Our results are obtained analytically and validated with simulated and empirical data. A software program is provided that clinics can readily use to solve the wait-preempt dilemma.

An important phase of a wind farm design is solving the Wind Farm Layout Optimization Problem (WFLOP), which consists in optimally positioning the turbines within the wind farm so that the wake effects are minimized and therefore the expected power production maximized. Although this problem has been receiving increasing attention from the scientific community, the existing approaches do not completely respond to the needs of a wind farm developer, mainly because they do not address construction and logistical issues. This chapter describes the WFLOP, gives an overview on the existing work, and discusses the challenges that may be overcome by future research.