Machine learning-based failure prediction in United States of America lobbying firms : the role of director networks

van der Haar, Dustin, Lodh, Suman and Nandy, Monomita (2025) Machine learning-based failure prediction in United States of America lobbying firms : the role of director networks. Engineering Applications of Artificial Intelligence, 156(B), p. 111091. ISSN (print) 0952-1976

Abstract

The role of directors and their associated social networks is critical in determining corporate failure, particularly within lobbying firms in the United States. However, predicting corporate failure in increasingly complex organizational structures remains an underexplored challenge. Using firm-level director data from the United States between 2005 and 2018, we apply both traditional machine learning models – Logistic Regression, Random Forest, and Support Vector Machines (SVM) – and more recent tabular deep learning approaches, including TabTransformer and Feature Tokenizer Transformer, to predict and uncover complex non-linear relationships between director network attributes and firm failure. Our results demonstrate that corporate failure can be accurately predicted based on director network characteristics, achieving a mean performance of 99.98% Area Under the Receiver Operating Characteristic Curve (AUC-ROC), 85.10% Precision, and 80.23% Recall using a stratified 5-fold cross-validation procedure with a weighted averaging strategy. Furthermore, network-derived attributes such as centrality, betweenness, and remuneration significantly influence failure risk in lobbying firms. These findings contribute to the management literature by highlighting the predictive value of director networks in corporate failure and offer practical insights for directors, engineering practitioners, and policymakers seeking to mitigate organizational risk.

Actions (Repository Editors)

Item Control Page Item Control Page