InsurePricing — Insurance Premium Pricing Models

Created by Bytical AI — AI agents that run insurance operations.

Model Description

InsurePricing provides two complementary pricing models for UK motor insurance premiums, designed for actuarial and underwriting workflows:

1. Tweedie GLM — Generalized Linear Model with Tweedie distribution (power=1.5), the industry-standard approach for insurance pricing
2. Explainable Boosting Machine (EBM) — Interpretable glass-box model from Microsoft Research (InterpretML) that provides per-feature explanations

Model Comparison

Model	MAE (£)	RMSE (£)	MAPE (%)	Interpretable
EBM	£11,132	£14,787	177.6%	Yes — per-feature shape functions
Tweedie GLM	£12,245	£17,615	198.8%	Yes — coefficients

Risk Factors (17 Features)

Feature	Type	Description
driver_age	Numeric	Age of primary driver
years_driving	Numeric	Years of driving experience
years_ncd	Numeric	No-claims discount years
vehicle_year	Numeric	Year of vehicle manufacture
vehicle_value	Numeric	Vehicle market value (£)
annual_mileage	Numeric	Estimated annual miles
voluntary_excess	Numeric	Voluntary excess amount (£)
compulsory_excess	Numeric	Compulsory excess amount (£)
previous_claims_3y	Numeric	Claims in last 3 years
policy_age_days	Numeric	Days since policy inception
vehicle_age	Derived	Current year minus vehicle_year
driver_experience_ratio	Derived	years_driving / driver_age
ncd_ratio	Derived	years_ncd / years_driving
vehicle_make_enc	Encoded	Vehicle manufacturer
fuel_type_enc	Encoded	Fuel type
occupation_enc	Encoded	Driver occupation
region_enc	Encoded	UK region

EBM Top Feature Importances

Feature	Importance
previous_claims_3y	3,259
policy_age_days	2,684
previous_claims_3y × policy_age_days	1,608
region_enc	221
vehicle_make_enc	173
annual_mileage	172
compulsory_excess	165
voluntary_excess	163
ncd_ratio	153

Training Data

• 25,000 synthetic UK motor insurance policies (20K train / 5K test)
• Features include driver demographics, vehicle attributes, claim history, and policy details

Files

File	Description
tweedie_glm.pkl	Scikit-learn Tweedie GLM pipeline
pricing_ebm.pkl	InterpretML EBM model
label_encoders.pkl	Fitted label encoders for categorical features
training_results.json	Full training metrics and feature coefficients

How to Use

import pickle
import numpy as np

# Load EBM model
with open("pricing_ebm.pkl", "rb") as f:
    ebm = pickle.load(f)
with open("label_encoders.pkl", "rb") as f:
    encoders = pickle.load(f)

# Example: price a motor policy
features = np.array([[
    30,     # driver_age
    8,      # years_driving
    4,      # years_ncd
    2022,   # vehicle_year
    20000,  # vehicle_value
    10000,  # annual_mileage
    200,    # voluntary_excess
    100,    # compulsory_excess
    0,      # previous_claims_3y
    180,    # policy_age_days
    4,      # vehicle_age
    0.267,  # driver_experience_ratio
    0.5,    # ncd_ratio
    3,      # vehicle_make_enc
    1,      # fuel_type_enc
    5,      # occupation_enc
    7       # region_enc
]])

premium = ebm.predict(features)[0]
print(f"Predicted premium: £{premium:,.2f}")

# Get per-feature explanations (EBM glass-box)
explanations = ebm.explain_local(features)

Part of the INSUREOS Model Suite

This model is part of the INSUREOS — a complete AI/ML suite for insurance operations built by Bytical AI:

Model	Task	Metric
InsureLLM-4B	Insurance domain LLM	ROUGE-1: 0.384
InsureDocClassifier	12-class document classification	F1: 1.0
InsureNER	13-entity Named Entity Recognition	F1: 1.0
InsureFraudNet	Fraud detection (Motor/Property/Liability)	AUC-ROC: 1.0
InsurePricing (this model)	Insurance pricing (GLM + EBM)	MAE: £11,132

Citation

@misc{bytical2026insurepricing,
  title={InsurePricing: Explainable Insurance Premium Pricing Models},
  author={Bytical AI},
  year={2026},
  url={https://huggingface.co/piyushptiwari/InsurePricing}
}

About Bytical AI

Bytical builds AI agents that run insurance operations — claims automation, underwriting intelligence, digital sales, and core system modernization for insurers across the UK and Europe. Microsoft AI Partner | NVIDIA | Salesforce.