Decision Support Systems最新文献

Shapley values have become one of the go-to methods to explain complex models to end-users. They provide a model agnostic post-hoc explanation with foundations in game theory: what is the worth of a player (in machine learning, a feature value) in the objective function (the output of the complex machine learning model). One downside is that they always require outputs of the model when some features are missing. These are usually computed by taking the expectation over the missing features. This however introduces a non-trivial choice: do we condition on the unknown features or not? In this paper we examine this question and claim that they represent two different explanations which are valid for different end-users: one that explains the model and one that explains the model combined with the feature dependencies in the data. We propose a new algorithmic approach to combine both explanations, removing the burden of choice and enhancing the explanatory power of Shapley values, and show that it achieves intuitive results on simple problems. We apply our method to two real-world datasets and discuss the explanations. Finally, we demonstrate how our method is either equivalent or superior to state-to-of-art Shapley value implementations while simultaneously allowing for increased insight into the model-data structure.

This study aims to assess the financial statement fraud risk ex ante and empirically explore its information content to help improve decision-making and daily operations. We propose an ex-ante fraud risk index by adopting an ensemble learning approach and a theoretically grounded framework. Our ensemble learning model systematically examines the fraud process and deals effectively with the unique challenges in the financial fraud setting, which yields superior prediction performance. More importantly, we empirically examine the information content of our estimated ex-ante fraud risk from the perspective of operational efficiency. Our empirical results find that the estimated ex-ante fraud risk is negatively correlated with sustaining operational efficiency. This study redefines fraud detection as an ongoing endeavor rather than a retrospective event, thus enabling managers and stakeholders to reconsider their operation decisions and reshape their entire operation processes accordingly.

Malicious content threatens the integrity and quality of content in social networks. Research and practice have experimented with network intervention strategies to curb malicious content propagation. These strategies lack efficiency, target malicious content propagators, and abridge freedom of speech. We draw upon the preferential attachment literature and cognitive load theory to employ the mechanisms of network formation, information sharing, and limited human cognitive capacities to propose an alternative feed management strategy—Preferentiality Dampened Feed Management. We compare and contrast this strategy against other established strategies using an agent-based model that utilizes empirical data from Twitter and findings from the prior literature. The results from our two experiments suggest that our proposed strategy is more effective in curbing malicious content propagation than other established strategies. Our work has important implications for the network interventions literature and practical implications for platform providers, social media users, and society.

Beyond managing student dropout, higher education stakeholders need decision support to consistently influence the student learning process to keep students motivated, engaged, and successful. At the course level, the combination of predictive analytics and self-regulation theory can help instructors determine the best study advice and allow learners to better self-regulate and determine how they want to learn. The best performing techniques are often black-box models that favor performance over interpretability and are heavily influenced by course contexts. In this study, we argue that explainable AI has the potential not only to uncover the reasons behind model decisions, but also to reveal their stability across contexts, effectively bridging the gap between predictive and explanatory learning analytics (LA). In contributing to decision support systems research, this study (1) leverages traditional techniques, such as concept drift and performance drift, to investigate the stability of student success prediction models over time; (2) uses Shapley Additive explanations in a novel way to explore the stability of extracted feature importance rankings generated for these models; (3) generates new insights that emerge from stable features across cohorts, enabling teachers to determine study advice. We believe this study makes a strong contribution to education research at large and expands the field of LA by augmenting the interpretability and explainability of prediction algorithms and ensuring their applicability in changing contexts.

With the development of e-commerce, collective overreactions such as buying frenzy have become prominent. However, studies have rarely investigated the mechanism of irrational consumer behavior at the group level. To investigate the evolution of collective overreaction in dynamic online product diffusion networks, we employed a sequential multiple-methods approach. A conceptual model is constructed to capture the influence of social network dynamic evolution on individual irrationality. An agent-based model (ABM) under different network dynamic growth mechanisms is implemented and verified. The findings revealed the following. In external dynamic growth mechanisms, key opinion consumer (KOC) connection can lead to positive collective overreaction (i.e., the adoption rate of consumer groups spikes). This effect fades as the probability of KOC connection increases and stabilizes as the node change rate decreases. Random connection is prone to negative collective overreaction (i.e., a sudden and sharp decline in the adoption rate of consumer groups), and key opinion leader (KOL) connection exhibits both positive and negative collective overreaction. Increasing the edge change rate increases the frequency of negative collective overreaction in KOL connections. In internal dynamic growth mechanisms, KOL and KOC connections are prone to negative collective overreaction; increasing the edge change rate can reduce the frequency of negative collective overreaction in KOL overreaction, and an appropriate edge change rate can inhibit the emergence of negative collective overreaction in KOC connection. This research contributes to the area of internet product marketing and provides a new basic framework through which to combine psychology and the ABM.

The field of decision sciences is undergoing significant disruption and reinvention because of rapid advances in artificial intelligence (AI) technologies and the design of complex human-artificial intelligence systems (HAIS). The integration of human decision behaviors with cutting-edge AI capabilities is transforming business and society in irreversible ways. In this paper, we examine prior research published in Decision Support Systems that makes contributions to HAIS design science research (DSR). We define synergistic interactions among DSR, AI technology design, and human interaction design, which we use to specify the dimensions for an analysis of the DSS HAIS literature. We identify key challenges, leading to future research directions for the design of HAIS as solutions for complex decision science problems.

Teams are building blocks of organizations and essential inputs of organizational success. This article studies a data-driven analytical approach that exploits the rich data accumulated in organizations in the digital era to design teams, including prescribing team composition and formation decisions. We propose to evaluate a team regarding its performance and temporal stability, referred to as sustainable effectiveness (SE). Our approach estimates the team's performance and stability using machine learning models. It then optimizes an integrated objective of the team's performance and stability through mixed-integer programming models formulated according to predictive models. Consequently, this approach mines meaningful team compositions from historical data and guides strategic team formation accordingly. We conduct empirical studies using authentic data from our partner company in the real estate brokerage industry. The findings reveal that teams that adhere to our model's recommendations achieve an average percentage improvement of 153.1% to 156.5% higher than the benchmark teams, particularly when recruiting one or two members in their actual SE during the post-formation period. We further disclose the mechanism underlying this improvement from the perspective of changes in team compositions. Our study provides a decision support tool for team design and ensuing team dynamic management.

Electronic health records (EHRs) are a valuable source of information that can aid in understanding a patient’s health condition and making informed healthcare decisions. However, modelling longitudinal EHRs with heterogeneous information is a challenging task. Although recurrent neural networks (RNNs) are frequently utilized in artificial intelligence (AI) models for capturing longitudinal data, their explanatory capabilities are limited. Predictive clustering stands as the most recent advancement within this domain, offering interpretable indications at the cluster level for predicting disease risk. Nonetheless, the challenge of determining the optimal number of clusters has put a brake on the widespread application of predictive clustering for disease risk prediction. In this paper, we introduce a novel non-parametric predictive clustering-based risk prediction model that integrates the Dirichlet Process Mixture Model (DPMM) with predictive clustering via neural networks. To enhance the model’s interpretability, we integrate attention mechanisms that enable the capture of local-level evidence in addition to the cluster-level evidence provided by predictive clustering. The outcome of this research is the development of a multi-level explainable artificial intelligence (AI) model. We evaluated the proposed model on two real-world datasets and demonstrated its effectiveness in capturing longitudinal EHR information for disease risk prediction. Moreover, the model successfully produced interpretable evidence to bolster its predictions.

Customer retention management relies on advanced analytics for decision making. Decision makers in this area require methods that are capable of accurately predicting which customers are likely to churn and that allow to discover drivers of customer churn. As a result, customer churn prediction models are frequently evaluated based on both their predictive performance and their capacity to extract meaningful insights from the models. In this paper, we extend hybrid segmented models for customer churn prediction by incorporating powerful models that can capture non-linearities. To ensure the interpretability of such segmented hybrid models, we introduce a novel model-agnostic approach that extends SHAP. We extensively benchmark the proposed methods on 14 customer churn datasets on their predictive performance. The interpretability aspect of the new model-agnostic approach for interpreting hybrid segmented models is illustrated using a case study. Our contributions to decision making literature are threefold. First, we introduce new hybrid segmented models as powerful tools for decision makers to boost predictive performance. Second, we provide insights in the relative predictive performance by an extensive benchmarking study that compares the new hybrid segmented methods with their base models and existing hybrid models. Third, we propose a model-agnostic tool for segmented hybrid models that provide decision makers with a tool to gain insights for any hybrid segmented model and illustrate it on a case study. Although we focus on customer retention management in this study, this paper is also relevant for decision makers that rely on predictive modeling for other tasks.

Hepatitis carcinoma (HCC) accounts for the majority of liver cancer–related deaths globally. Cirrhosis often precedes HCC clinically in a strong, temporal relationship. Therefore, identifying cirrhosis patients at higher risk of HCC is crucial to physicians' clinical decision-making and patient management. Effective estimates of at-risk patients can facilitate timely therapeutic interventions and thereby enhance patient outcomes and well-being. We develop a novel, meta-path, attention-based deep learning method to identify at-risk cirrhosis patients. The proposed method integrates complex patient–medication interactions, essential patient–patient and medication–medication links, and the combined effects of medication and comorbidity to support downstream predictions. An empirical test of the proposed method's predictive utilities, relative to nine existing methods, uses a large sample of real-world cirrhosis patient data. The comparative results indicate that the proposed method can identify at-risk patients more effectively than all the benchmarks. The current research has important implications for clinical decision support and patient management, and it can facilitate patient self-management and treatment compliance too.