Information Systems Faculty Publications and Presentations

Document Type

Article

Publication Date

11-10-2023

Abstract

Highlights

  • Holistic XAN model combines descriptive, predictive, prescriptive analytics.

  • Cutting-edge techniques for feature selection, optimization, and explanations.

  • Transparent justifications for factors enhance trust, and assist domain experts.

  • Interpretable representations assist in intelligent decision-making.

Abstract

Most analytics models are built on complex internal learning processes and calculations, which might be unintuitive, opaque, and incomprehensible to humans. Analytics-based decisions must be transparent and intuitive to foster greater human acceptability and confidence in analytics. Explainable analytics models are transparent models in which the primary factors and weights that lead to a prediction can be explained. Typical AI models are non-transparent or opaque models, in which even the designers cannot explain how their models arrive at a specific decision. These transparent models help decision-makers understand their judgments and build trust in analytics. This study introduces an innovative, comprehensive model that fuses descriptive, predictive, and prescriptive analytics, offering a fresh perspective on car accident severity. Our methodological contribution lies in the application of advanced techniques to address data-related challenges, optimize feature selection, develop predictive models, and fine-tune parameters. Importantly, we also incorporate model-agnostic interpretation techniques, further enhancing the transparency and interpretability of our model, and separate explanations from models (i.e., model-agnostic interpretation techniques). Our findings should provide novel insights for a domain expert to understand accident severity. The explainable analytics approach suggested in this study supplements non-transparent machine learning prediction models, particularly optimized ensemble models. Our model's end product is a comprehensible representation of crash severity factors. To obtain a more trustworthy assessment of accident severity, this model may be supplemented with insurance data, medical data such as blood work and pulse rate, and previous medical history.

Comments

Original published version available at https://doi.org/10.1016/j.ejor.2023.11.013

Publication Title

European Journal of Operational Research

DOI

10.1016/j.ejor.2023.11.013

Download

Available for download on Monday, November 10, 2025

Included in

Business Commons

Share

COinS