
Emre Saylan
Enchancing Credit Insurance with Glassbox Models and LLMs for Transparent Decision-Making.
Rel. Daniele Apiletti. Politecnico di Torino, Corso di laurea magistrale in Data Science And Engineering, 2024
Abstract: |
Credit insurance is a critical component of the financial ecosystem, protecting busi- nesses against the risk of non-payment for goods and services provided on credit. This insurance promotes economic stability by allowing companies to extend credit with re- duced risk, thus supporting business growth and resilience. This thesis, conducted in collaboration with Allianz Trade the sector leader in credit insurance focuses on develop- ing an explainable artificial intelligence (AI) framework to automate and enhance credit limit decision-making within the industry. The thesis centers on building a transparent AI system that meets both predictive accuracy and interpretability requirements essential to credit insurance. To achieve this, an Explainable Boosting Machine (EBM) was designed and trained on a comprehen- sive dataset of approximately three million credit limit requests across multiple Euro- pean countries and industry sectors. The EBM model’s interpretable structure provides insights into feature importance and non-linear relationships, ensuring that the credit decision-making process remains understandable and aligned with regulatory standards. To facilitate clear communication of model outputs, particularly in partially approved or rejected cases, a Large Language Model (LLM) was integrated to generate human- readable justifications. Utilizing Anthropic’s Claude LLM, the system employs advanced prompt engineering techniques, such as Chain of Thought (CoT) reasoning and a priority- based motive structure, to create explanations that align with the underwriting team’s priorities. This enables the model to produce coherent, non-technical explanations. After these non-technical explanations automatically converted to professional word document that credit underwriting team can benefit. This thesis contributes a robust, interpretable AI framework tailored for high-stakes decision-making in credit insurance, with potential applications in other fields where transparency and accountability are essential. |
---|---|
Relators: | Daniele Apiletti |
Academic year: | 2024/25 |
Publication type: | Electronic |
Number of Pages: | 66 |
Additional Information: | Tesi secretata. Fulltext non presente |
Subjects: | |
Corso di laurea: | Corso di laurea magistrale in Data Science And Engineering |
Classe di laurea: | New organization > Master science > LM-32 - COMPUTER SYSTEMS ENGINEERING |
Aziende collaboratrici: | EULER HERMES SERVICES |
URI: | http://webthesis.biblio.polito.it/id/eprint/33784 |
![]() |
Modify record (reserved for operators) |