[Paper Review] Using Decision Tree Learner to Classify Solvency Position for Thai Non-life Insurance Companies

Phaiboon Jhongpita, Sukree Sinthupinyo+1 more|arXiv (Cornell University)|Mar 14, 2012

Imbalanced Data Classification Techniques10 references3 citations

TL;DR

This study proposes a decision tree learner model to classify the solvency position of Thai non-life insurance companies into four categories: strong, moderate, weak, or insolvent. Using 10-fold cross-validation, train-test split, and separate test sets, the model achieves high accuracy in early warning classification, demonstrating strong predictive performance for financial stability assessment in the non-life insurance sector.

ABSTRACT

This paper introduces a Decision Tree Learner as an early warning system for classification of the non-life insurance companies according to their financial solid as strong, moderate, weak, or insolvency. In this study, we ran several experiments to show that the proposed model can achieve a good result using standard 10 fold crossvalidation, split train and test data set, and separated test set. The results show that the method is effective and can accurately classify the solvency position.

Motivation & Objective

• To develop an early warning system for identifying financial solvency status in Thai non-life insurance companies.
• To classify solvency into four distinct categories: strong, moderate, weak, or insolvency.
• To evaluate the performance of a decision tree learner model using multiple validation techniques.
• To provide a data-driven, interpretable tool for regulators and insurers to monitor financial health.
• To assess the model’s robustness across different data partitioning strategies.

Proposed method

• A decision tree learner algorithm is applied to financial data from Thai non-life insurance companies.
• The model uses 10-fold cross-validation to assess generalization performance.
• The dataset is split into training and testing sets to validate model accuracy.
• A separate, independent test set is used to evaluate final model performance.
• Feature selection is implicitly handled by the decision tree algorithm through recursive partitioning based on information gain.
• The model classifies companies into four solvency categories based on financial ratios and performance indicators.

Experimental results

Research questions

• RQ1 Can a decision tree learner accurately classify the solvency status of Thai non-life insurance companies?
• RQ2 How does the model perform across different data partitioning methods such as cross-validation and separate test sets?
• RQ3 What is the predictive accuracy of the decision tree model in distinguishing between strong, moderate, weak, and insolvent financial positions?
• RQ4 Is the decision tree model robust and reliable as an early warning system for solvency risk?
• RQ5 How interpretable and actionable is the resulting decision tree for financial regulators and insurers?

Key findings

• The decision tree learner model achieved high classification accuracy in predicting solvency positions across multiple evaluation methods.
• The model demonstrated consistent performance using 10-fold cross-validation, train-test split, and a separate test set.
• The results confirm that the decision tree approach is effective for early warning classification in non-life insurance solvency monitoring.
• The model successfully categorized companies into four solvency levels: strong, moderate, weak, and insolvency.
• The study confirms the feasibility of using interpretable machine learning models for financial risk assessment in the insurance sector.
• The model's performance supports its potential use as a practical tool for regulatory and internal financial monitoring.