Managing Humanitarian Access and Information Gaps: A Deep Learning Approach with MLP and Permutation Analysis

Abstract This article introduces a comprehensive framework to explore the relationship between access constraints and information deficits in humanitarian logistics. By integrating resource dependence theory (RDT) and institutional theory (IST), we examined how barriers such as movement restrictions, violence, and the presence of mines impact resource flow and operational efficiency. Drawing on RDT and IST, we analyzed data from 150 crisis events across 93 countries, encompassing nine indicators of humanitarian access constraints and information gaps. Utilizing advanced deep learning techniques, including the multi-layer perceptron (MLP) model with permutation feature importance analysis, our findings reveal that restrictions on movement within countries, violence against personnel and facilities, and administrative obstructions are the most significant predictors of information gaps. These constraints significantly hinder humanitarian responses by disrupting resource distribution and violating institutional norms. The findings highlight the primacy of access-related barriers in undermining timely and accurate needs assessment, providing empirical support for the theoretical integration of RDT and IST. The framework offers actionable insights into crisis information dynamics and informs more targeted and resilient response strategies.