A Novel Hybrid Model for Loan Default Prediction in Maritime Finance Based on Topological Data Analysis and Machine Learning

The global shipping industry, pivotal to international trade, faces inherent financial vulnerabilities due to cyclical asset values, volatile freight rates, and high leverage, rendering traditional credit risk models inadequate. This study proposes a novel framework integrating topological data analysis (TDA) and machine learning (ML) to enhance default prediction in maritime finance. By constructing correlation-based networks of shipping firms and extracting topological persistence features, such as cyclical trends and structural interdependencies, via Vietoris-Rips complexes, the model captures nonlinear risk patterns overlooked by conventional metrics. Network properties (e.g., centrality, clustering) and financial indicators (loan terms, vessel attributes, economic indices) are synthesized, enabling Graph Neural Networks (GNNs) to leverage relational and topological insights. Evaluated on a dataset of shipping loans spanning financial, operational, and macroeconomic variables, the TDA-enhanced models demonstrate significant improvements: logistic regression accuracy rose from 62.8% to 85.1%. At the same time, ROC-AUC for SVM and XGBoost reached 0.931 and 0.973, respectively. Key features like loan amount, vessel age, and debt-to-equity ratios exhibited strong correlations with defaults. Results underscore that incorporating TDA-derived persistence features improves early detection of financial distress, particularly in capturing systemic risks from interconnected market shocks. This hybrid approach offers a robust tool for lenders and policymakers to mitigate defaults in a sector critical to global commerce.