What question did this study set out to answer?

This research aims to enhance loan default prediction in maritime finance using advanced analytical methods.

April 27, 2026Open Access

Predicting Loan Defaults in Maritime Finance Through Topological Data Analysis

JSJinka Shiva GVGunti ViswanathJawaharlal Nehru Technological University Anantapur BPBoya Shiva Prasad

Key Points

This research aims to enhance loan default prediction in maritime finance using advanced analytical methods.
Developed a correlation-based network using Topological Data Analysis and machine learning techniques.
Utilized financial, operational, and topological data for training four classifiers: Logistic Regression, Decision Tree, Random Forest, and XGBoost.
Evaluated model performance using accuracy and F1-Score metrics.
XGBoost achieved an accuracy of 90.00% and an F1-Score of 91.43%, outperforming other classifiers.
Incorporating topological features improved prediction results for all classifiers.
Network data provided additional predictive value beyond standard financial metrics.

Abstract

The maritime finance industry struggles to forecast loan defaults because its highly unpredictable markets and changing freight rates and interconnected risk systems.The tra-ditional credit risk assessment methods for shipping companies use financial ratios and historical performance data to evaluate companies as independent entities, which restricts their ability to understand the intricate relationship network that exists among companies operating in identical market conditions and trade routes.The research presents a combined method that merges Topological Data Analysis with machine learning techniques to enhance the precision of loan default predictions in maritime finance.The framework develops a correlation-based network in which shipping companies function as nodes and their financial and operational connections determine the weight of their inter-company links.The network generates network-based topological metrics that include degree centrality, betweenness centrality, clustering coefficient, and PageRank which help to identify structural patterns and systemic interdependencies that standard financial data cannot reveal.The Pearson correlation coefficient identifies the most important features for predicting loan default through its analysis of the linear relationship strength between each input feature and the Default target variable.The feature set contains financial, operational, and topological data which en-ables four machine learning classifiers to be trained on the dataset that includes Logistic Regression, Decision Tree, Random Forest, and XGBoost.The experimental results show that XGBoost delivers the best classification results which include an accuracy of 0.9000 and an F1-Score of 0.9143, because it outperforms all other models in the baseline configuration.The topological features which we added to the model showed better predic-tion results with all four classifiers, because the network data which includes relational and structural information provided additional predictive value beyond standard financial metrics.The new hybrid system which we developed enables financial institutions to assess credit risk in the maritime shipping industry through a complete, precise, and operational assessment tool.

KI fragen

Bookmark

View Full Paper