PlumBot: Bot: Creating new article from JSON

2026-03-11T16:58:48Z

Bot: Creating new article from JSON

New page

📋 '''Decision tree''' is a supervised [[Definition:Machine learning | machine-learning]] and statistical modeling technique that segments data into branches based on a sequence of conditional rules, producing a tree-like structure that maps inputs to predicted outcomes. In insurance, decision trees are deployed across [[Definition:Underwriting | underwriting]], [[Definition:Claims management | claims triage]], [[Definition:Fraud detection | fraud detection]], and [[Definition:Pricing | pricing]] workflows because they deliver predictions that are not only accurate but also interpretable—a critical requirement in a heavily regulated industry where [[Definition:Insurance regulator | regulators]] may demand explanations for adverse decisions affecting [[Definition:Policyholder | policyholders]].

⚙️ A typical application begins with a training dataset—historical [[Definition:Policy | policies]] labeled with outcomes such as [[Definition:Claim | claim]] occurrence or [[Definition:Loss ratio (L/R) | loss-ratio]] bands. The algorithm evaluates each available feature (e.g., driver age, property location, prior [[Definition:Claims history | claims history]]) and selects the split at each node that maximizes information gain or minimizes impurity. The resulting model can, for example, route incoming [[Definition:Submission | submissions]] in an [[Definition:Managing general agent (MGA) | MGA's]] pipeline: high-confidence risks flow straight to [[Definition:Binding authority agreement | bind]], borderline cases get flagged for human review, and clearly out-of-appetite risks receive an automated decline. Ensemble methods like [[Definition:Random forest | random forests]] and [[Definition:Gradient boosting | gradient-boosted trees]] stack multiple decision trees to improve predictive power while reducing [[Definition:Overfitting | overfitting]], and they now form the backbone of many [[Definition:Insurtech | insurtech]] rating engines.

🔎 What makes decision trees especially valuable in insurance is the transparency of their logic. Unlike deep [[Definition:Neural network | neural networks]], a single decision tree can be visualized as a flowchart, enabling [[Definition:Actuary | actuaries]] and [[Definition:Compliance | compliance]] officers to trace exactly why a particular risk was classified the way it was. This auditability helps carriers satisfy [[Definition:Fair lending | fair-pricing]] and [[Definition:Anti-discrimination | anti-discrimination]] requirements, since each branching criterion can be reviewed for unintended proxy effects on protected classes. As the industry moves toward greater reliance on [[Definition:Artificial intelligence (AI) | artificial intelligence]], the decision tree remains a foundational building block—valued both as a standalone tool and as a component inside more complex [[Definition:Predictive model | predictive models]].

'''Related concepts'''
{{Div col|colwidth=20em}}
* [[Definition:Machine learning]]
* [[Definition:Random forest]]
* [[Definition:Predictive model]]
* [[Definition:Fraud detection]]
* [[Definition:Underwriting]]
* [[Definition:Gradient boosting]]
{{Div col end}}

Definition:Decision tree - Revision history

PlumBot: Bot: Creating new article from JSON