Document Type : Original Research Paper

Authors

• Zeinab Jafarzadeh ¹
• Narges Mirehi ²
• Effat Golpar Raboky ³

¹ B.Sc. Student, Department of Computer Science, Faculty of Basic Sciences, University of Qom, Qom, Iran.

² Assistant Professor, Department of Computer Sciences, Faculty of Sciences, University of Qom, Qom, Iran.

³ Associate Professor, Department of Mathematical Sciences, Faculty of Sciences, University of Qom, Qom, Iran.

Abstract

BACKGROUND AND OBJECTIVES: Intelligent transportation and logistics management are critical components of modern supply chains, directly influencing cost reduction and risk management. With the rapid increase in transportation data, the need for data-driven approaches to financial decision-making has become increasingly evident. One of the main challenges in this field is the non-linear nature of relationships in premium calculations, which traditional models fail to adequately capture. This research aims to transition from conventional methods to an intelligent insurance pricing system in Qom Province, Iran—recognized as a major transit hub and the logistics center of the country. The significance of this study lies in its ability to uncover non-linear patterns within large-scale datasets that traditional models are unable to analyze effectively. The operational benefits of this system go beyond mere numerical predictions; unlike them, it includes real-time financial risk management, increased tariff transparency, and the intelligent detection of anomalies in the bill of lading issuance process. These features contribute to optimized budgeting and improved decision-making accuracy in this critical transportation bottleneck.
METHODS: This data-driven research focuses on over 600,000 bills of lading records from Qom Province. In the first stage, the data underwent thorough cleaning and outlier management, followed by the identification of key variables that influence insurance premiums such as cargo weight, transportation distance, and product type. Due to the high complexity and non-linear relationships among these variables, traditional linear models were unable to capture the underlying pricing patterns. To address this limitation, advanced algorithms including Symmetric Gradient Boosting (Categorical Boosting) and Extreme Gradient Boosting (XGBoost) were deployed. The Symmetric Gradient Boosting model was chosen as the core algorithm due to its superior ability to handle categorical variables (such as cargo type) and its robust performance in preventing overfitting. In the final stage, the models’ performance was assessed using statistical metrics, including the Coefficient of Determination (R²) and Median Absolute Error (MedAE), to evaluate the precision of the predicted values.
FINDINGS: The analysis revealed that intelligent algorithms significantly improve the accuracy of insurance premium calculations. The primary innovation of this research is the identification of hidden non-linear and multi-dimensional patterns within large-scale logistics data. The Symmetric Gradient Boosting model achieved an R² score of 0.86, accurately reconstructing market fluctuations with high precision. This result suggests that the model can replace traditional methods, which, due to their rigid structure, lack the flexibility to adapt to the complexities of the logistics industry. By leveraging advanced machine learning techniques, this study demonstrated that insurance pricing is governed by non-linear relationships rather than simple linear dependencies. Furthermore, the research showed that qualitative factors such as cargo type and the identity of the transportation companies play a far more significant role in financial risk management than before. Traditional models typically placed heavy emphasis on quantitative variables such as distance, but this study demonstrated that categorical variables could influence pricing much more significantly.
CONCLUSION: A key strategic takeaway from this study is the identification of cargo nature and the identity of transportation companies as the primary factors influencing insurance risk assessment. Unlike traditional methods, which predominantly focus on distance, the findings from this research suggest that qualitative parameters such as the type of goods being transported and the reliability of the transport company play a far more substantial role in managing financial risk. The intelligent architecture developed in this study minimizes prediction errors, making it possible to implement dynamic and fair pricing mechanisms. This not only ensures the protection of the interests of insurance companies but also safeguards the rights of cargo owners, thereby creating a more balanced and equitable system for all stakeholders. By moving beyond outdated pricing models, this intelligent system provides a foundation for a more transparent, efficient, and responsive insurance pricing model in the transportation and logistics sector. Moreover, the proposed system offers real-time capabilities for financial risk management and anomaly detection, enhancing operational efficiency and reducing human error. This is particularly crucial in sectors where financial decision-making is complex and data-heavy, such as logistics and transportation. The system also facilitates the detection of potential discrepancies in bill of lading issuance, thereby reducing fraudulent activities and ensuring compliance with industry standards. These capabilities can have a far-reaching impact on the logistics sector, improving the transparency and accountability of transactions while also supporting better budgeting and forecasting.

Keywords

• Anomaly detection
• Financial risk management
• Insurance premium prediction
• CatBoost Model
• Intelligent logistics

Main Subjects

• Risk management in the insurance industry