Machine Learning For An Explainable Cost Prediction of Medical Insurance

Explainability matters for insurance decisions: The study demonstrates that machine learning models used to predict medical insurance costs can be made transparent through explainable AI (XAI) methods like SHAP and Individual Conditional Expectation plots, allowing stakeholders to understand which factors drive premium prices rather than treating predictions as "black boxes."

Trade-offs between accuracy and efficiency exist: While the XGBoost model achieved the best predictive performance, the Random Forest model produced nearly comparable results while using significantly fewer computational resources, suggesting that "best" performance isn't always the most responsible choice when considering environmental and operational costs.

Multiple methods can validate findings: The study found that two different explainability techniques (SHAP and Individual Conditional Expectation plots) produced similar results in identifying key premium determinants, which increases confidence that the identified factors are genuinely important rather than artifacts of a single method.

Transparency enables fairness assessment: By identifying which specific factors most influence insurance cost predictions, this approach allows companies and regulators to examine whether those factors are appropriate, potentially helping detect and prevent discriminatory pricing practices in medical insurance.

Responsible AI requires multiple considerations: The research illustrates that responsible deployment of machine learning in insurance requires balancing multiple factors—accuracy, computational efficiency, explainability, and fairness—rather than optimizing for any single metric.