Analyse
Predicting Accident Risk for Future Policyholders
Trained a Random Forest model on 500,000 auto insurance policyholders enriched with external data, and delivered a claim-risk simulator with adjustable insurance premiums. Winning project of the Business Deep Dive.
Interactive demo
The claim-risk & pricing simulator
Describe a prospective policyholder and their vehicle: the model estimates the probability of a claim within the year, explains what drives it, and turns it into a premium you can adjust.
Loading the demo…
The challenge
Conducted during a three-week Business Deep Dive for Generali, the challenge centered on accurately assessing the claim risk of prospective auto insurance policyholders and translating model predictions into a practical pricing tool.
The question
How can historical data on 500,000 policyholders and external variables be leveraged to predict whether a new driver will file a claim within a year — and dynamically adjust their premium?
What I did
End-to-end data cleaning, feature engineering with external variables, and customer segmentation via clustering. Trained and tuned a Random Forest classifier evaluated across ROC-AUC, Accuracy and Precision metrics, then wrapped the model into an interactive simulator.
Backstage · Data sources
A dataset of 500,000 auto insurance policyholders covering driver characteristics and vehicle attributes, enriched with external contextual datasets.
The result
The team won the Business Deep Dive. Delivered a working simulator that predicts the probability of a claim for any prospective policyholder and computes adjustable insurance premiums based on predicted risk.