LINDA-BN: An interpretable probabilistic approach for demystifying black-box predictive models

2021 | Decision Support Systems | Citations: 0

Authors: Moreira, Catarina; Chou, Yu-Liang; Velmurugan, Mythreyi; Ouyang, Chun; Sindhgatta, Renuka; Bruza, Peter

Abstract: The use of sophisticated machine learning models for critical decision-making fa ... Expand

Abstract: The use of sophisticated machine learning models for critical decision-making faces the challenge that these models are often applied as a ‘black-box’. This has led to an increased interest in interpretable machine learning, where post-hoc model-agnostic algorithms present a useful mechanism for generating interpretations of complex learning models. This paper proposes a novel approach based on Bayesian Networks to generate local post-hoc model-agnostic interpretations of a black-box predictive model. Consequently, the proposed approach presents features that are conditionally dependent between each other and that are directly influencing the class variable. This enables the decision-maker to better understand how features are related and why a certain prediction was made. Compared to the existing post-hoc interpretation methods, the contribution of our approach is three-fold: (1) as a probabilistic graphical model, the extracted Bayesian network can provide interpretations through conditional dependencies in a graphical structure regarding what input features and how/why they contributed to a prediction; (2) for complex decision problems with many features, a Markov blanket can be generated from the extracted Bayesian network to provide interpretations with a focused view on those input features that directly contributed to a prediction; (3) the extracted Bayesian network enables the identification of four different rules which can inform the decision-maker about the confidence level in a prediction, thus helping the decision-maker assess the reliability of predictions learned by a black-box model. We implemented the proposed approach, applied it in the context of two well-known public datasets and analysed the results, which are made available in an open-source repository: https://github.com/catarina-moreira/LINDA_DSS. Collapse

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

Methods: Bayesian network computational algorithm experiment decision tree classification deep learning

Theories: game theory

“Alexa, Buy Me a Movie!”: How AI Speakers Reshape Digital Content Consumption and Preference

2018 | International Conference on Information Systems | Citations: 0

Authors: Son, Yoonseock; Oh, Wonseok

Abstract: Technologies based on artificial intelligence (AI) have revolutionized how indiv ... Expand

Abstract: Technologies based on artificial intelligence (AI) have revolutionized how individuals use various services by enabling verbal command over machines. Although such alternative forms of human-machine interaction have begun to insinuate themselves into our lives, limited efforts have been directed toward examining their economic effects. This study takes off from the current state of scholarship by examining the effects of AI speaker usage on consumers’ content consumption and preference behaviors. To this end, we used the dual processing system and self-control dilemma frameworks to analyze a unique panel data set that comprises information on household patterns of digital content transaction, consumption, and AI usage. Our results indicate that the adoption of AI speakers is positively associated with increased digital content purchase but negatively related to the average rate of content completion. We also find a change in type of content purchased, with customers consuming a greater proportion of hedonic products. The findings on households’ AI usage patterns indicate that emotion-seeking interactions are more positively associated with purchase than information-seeking interactions. On the basis of our findings, we drew implications for the design of optimal voice commerce strategies and derive insights into effectively eliciting the economic value of the voice recognition market. Collapse

Topics: smart speaker artificial intelligence consumer behavior human computer interaction usability

Methods: experimental group robustness check longitudinal research vector autoregression survey design

Theories: general systems theory

ENERGY INFORMATICS FOR ENVIRONMENTAL, ECONOMIC AND SOCIETAL SUSTAINABILITY: A CASE OF THE LARGE-SCALE DETECTION OF HOUSEHOLDS WITH OLD HEATING SYSTEMS

2016 | European Conference On Information Systems | Citations: 0

Authors: Kozlovskiy, Ilya; Sodenkamp, Mariya

Abstract: Energy utility companies hold a great potential to improve sustainable practices ... Expand

Abstract: Energy utility companies hold a great potential to improve sustainable practices, as they have access to large amounts of energy consumption data that can be translated into actionable insights toward individual consumers. We investigate the role of energy informatics based artifacts for utility companies to (i) profile households and to individually address customers that have a large potential to elevate their energy efficiency and (ii) at the same time create revenue by purchasing energy efficient goods and related services. Furthermore, we demonstrate how the predicted attributes of millions of households can, in turn, contribute to the economic, environmental, and societal value creation for energy suppliers, utility customers, and the community as a whole. For our empirical study on the detection of households with old heating systems, we cooperated with a large energy supplier in Belgium. The results are used for targeted positioning of new heating systems in a cross-selling campaign and may serve as a blueprint for related data analytics artifacts in this field. Collapse

Topics: energy information system marketing management cost benefit analysis value creation green IS

Methods: design artifact machine learning case study computational algorithm random forest classification

Machine learning and genetic algorithms in pharmaceutical development and manufacturing processes

2009 | Decision Support Systems | Citations: 1

Authors: Chi, Hoi-Ming; Moskowitz, Herbert; Ersoy, Okan K.; Altinkemer, Kemal; Gavin, Peter F.; Huff, Bret E.; Olsen, Bernard A.

Abstract: We develop an adaptive Automated Intelligent Manufacturing System (AIMS) for Cas ... Expand

Abstract: We develop an adaptive Automated Intelligent Manufacturing System (AIMS) for Case 1:to a well-understood-pharmaceutical-process to demonstrate our methodology, Case 2:with clustering, to a not-well-controlled or understood-process for seemingly identical experiments yielding disparate results, Case 3:to scale-up a process from development to manufacturing, and Case 4:to deploy AIMS adaptively, to modify the process model and reoptimize the system contemporaneously, when predictive errors are significant. The results showed AIMS had both explanatory and predictive power. We have developed the following methodological extensions: a random probe method for feature selection, a simulation approach to establish tolerances for target inputs, and an adaptive capability integrated with statistical-process-control to modify the model. Collapse

Topics: statistical process control new product development decision making manufacturing information system database system

Methods: computational algorithm support vector machine machine learning experiment cluster analysis

Theories: statistical learning theory

The Iterated Prisoner's Dilemma: early experiences with Learning Classifier System-based simple agents

2001 | Decision Support Systems | Citations: 0

Authors: Meng, Chen-Lu; Pakath, Ramakrishnan

Abstract: Prior research on artificial agents/agencies involves entities using specificall ... Expand

Abstract: Prior research on artificial agents/agencies involves entities using specifically tailored operational strategies (e.g., for information retrieval, purchase negotiation). In some situations, however, an agent must interact with others whose strategies are initially unknown and whose interests may counter its own. In such circumstances, pre-defining effective counter-strategies could become difficult or impractical. One solution, which may be viable in certain contexts, is to create agents that self-evolve increasingly effective strategies from rudimentary beginnings, during actual deployment. Using the Iterated Prisoner's Dilemma (IPD) problem as a generic agent-interaction setting, we use the Learning Classifier System (LCS) paradigm to construct autonomously adapting “simple” agents. A simple agent attempts to cope by maintaining an evolving but potentially perennially incomplete and imperfect knowledge base. These agents operate against specifically tailored (non-adaptive) agents. We present a preliminary suite of simulation experiments and results. The promise evidenced leads us to articulate several additional areas of interesting investigations that we are pursuing. Collapse

Topics: team productivity knowledge base knowledge representation artificial intelligence

Methods: experiment simulation experimental design machine learning application

Theories: prisoner's dilemma