همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Beyond the Victors: Correcting Survivorship Bias in Organizational Analytics with Explainable AI

نویسندگان :

Seyed Mohsen Tofighi¹ Hasan Pouladi²

1- Islamic Azad University, North Tehran Brancch 2- Islamic Azad University, Science and Research Brancch

کلمات کلیدی :

Survivorship Bias،Explainable AI (XAI)،Organizational Learning،Algorithmic Fairness،Insurance Industry

چکیده :

Organizational analytics frequently falls prey to survivorship bias, a cognitive pitfall where analysis focuses on successful entities “survivors” while neglecting unsuccessful ones. This practice yields distorted insights and flawed managerial strategies, particularly in high-turnover sectors like the insurance industry. This paper introduces the Analytical Debiasing Framework (ADF), a novel approach that integrates cognitive decision theory with Explainable Artificial Intelligence (XAI) to identify and correct this bias. Analyzing a dataset of 1,000 insurance agents (2020–2024), encompassing both retained and departed employees, we employ an XGBoost model enhanced with SHAP (SHapley Additive exPlanations). Our findings demonstrate that models trained exclusively on “survivors” mask critical performance drivers and suffer from lower predictive power. Conversely, a holistic model incorporating data from all agents achieves superior accuracy (AUC = 0.91) and reveals that organizational support mechanisms, often invisible in survivor-only datasets, are paramount for retention. By making the “reasons behind the predictions” transparent, the ADF transforms predictive models from black-box tools into cognitive instruments for organizational learning, fostering more robust, ethical, and effective decision-making.