Insurance Claims Prediction — Canadian P&C Market

Author: Reda Hakkani | PhD Candidate, Applied Mathematics | Montréal, QC
Domain: Machine Learning · Actuarial Pricing · Canadian P&C Insurance
Regulatory context: FSRA (ON) · AMF (QC) · AIRB (AB) · OSFI E-23 · IBC CANATICS

---

Overview

Full end-to-end ML pipeline for binary claims prediction on Canadian personal auto and commercial lines.
Calibrated to IBC industry statistics across 6 provinces: Ontario, Quebec, Alberta, BC, Manitoba, Saskatchewan.

Pipeline: EDA → Feature Engineering → GLM Baseline → XGBoost → Isotonic Calibration → SHAP

---

Results

Metric	GLM Baseline	XGBoost	Calibrated XGBoost
AUC-ROC	0.569	0.563	0.641
Gini Normalized	0.138	0.126	0.282
Decile 1 Lift	1.3x	1.5x	3.1x

---

Canadian Market Features

Dataset — 595,212 Policies

Feature	Description	Regulatory Relevance
province	ON / QC / AB / BC / MB / SK	Territory rating (FSRA/AMF)
vehicle_age	0–3 categories	IBC actuarial tables
driver_age_grp	5 age bands	FSRA rating factor
prior_claims	Claims history	Experience rating
canatics_flag	IBC fraud network flag	CANATICS integration
avg_daily_km	Telematics UBI	Intact / Desjardins PAYD
night_driving	% nocturnal	UBI risk indicator
hard_braking	Events / month	Telematics scoring
at_fault	Fault determination	AB / ON direct compensation

Engineered Features

'young_at_fault'         # High-risk interaction (driver_age_grp=1 × at_fault)
'fraud_signal'           # CANATICS flag × no witnesses
'high_risk_telematics'   # Hard braking OR speeding composite
'repeat_claimant'        # Prior claims ≥ 2
'cost_premium_ratio'     # Total claim cost / annual premium
'night_hard_interaction' # Night driving × hard braking events

---

Pipeline Architecture

595,212 Canadian P&C Policies
           │
           ▼
    ┌─────────────────────┐
    │   EDA & Quality     │
    │ - Province breakdown│
    │ - Claim rate: 5.1%  │
    │ - Class imbalance   │
    └──────────┬──────────┘
               │
               ▼
    ┌─────────────────────┐
    │ Feature Engineering │
    │ 21 → 31 features    │
    │ UBI + CANATICS +    │
    │ Interaction terms   │
    └──────────┬──────────┘
               │
         ┌─────┴──────┐
         ▼            ▼
    ┌─────────┐  ┌───────────────┐
    │   GLM   │  │  XGBoost      │
    │Baseline │  │ 250 estimators│
    │AUC:0.569│  │ AUC: 0.563    │
    └─────────┘  └──────┬────────┘
                        │
                        ▼
               ┌─────────────────┐
               │   Isotonic      │
               │  Calibration    │
               │  AUC: 0.641     │
               └────────┬────────┘
                        │
                        ▼
               ┌─────────────────┐
               │  SHAP Values    │
               │ OSFI E-23 ready │
               │ Underwriting UI │
               └─────────────────┘

---

Claim Rate by Province

Province	Policies	Claim Rate
Ontario	226,110	5.10%
Quebec	142,841	5.29%
Alberta	95,617	5.26%
British Columbia	83,104	4.76%
Manitoba	23,743	5.08%
Saskatchewan	11,875	5.15%

---

Top Predictors (Feature Importance)

Rank	Feature	Importance	Canadian Context
1	vehicle_age	0.115	IBC actuarial classification
2	night_driving	0.110	UBI / PAYD telematics
3	prior_claims	0.092	Experience rating (FSRA)
4	night_hard_interaction	0.091	Composite risk indicator
5	repair_cost	0.075	AB/ON direct compensation
6	speeding_pct	0.056	Telematics UBI

---

Regulatory Alignment

Standard	Application	Status
FSRA (ON)	Rate filing support	✓ Territory + experience factors
AMF (QC)	Tarification auto	✓ Territory pricing
AIRB (AB)	Grid rating system	✓ Vehicle class alignment
OSFI E-23	Model risk management	✓ SHAP interpretability
IBC CANATICS	Fraud detection network	✓ Flag integrated

---

Installation

git clone https://github.com/RedaHakkani/insurance-claims-prediction.git
cd insurance-claims-prediction
pip install -r requirements.txt
python src/claims_prediction.py

Requirements

numpy>=1.24.0
pandas>=2.0.0
scikit-learn>=1.3.0
matplotlib>=3.7.0

Output

• Full console report (province breakdown, lift table, feature importance)
• claims_prediction_results.png — 6-panel dashboard

---

References

• IBC (2023). Facts of the General Insurance Industry in Canada.
• FSRA (2022). Ontario Automobile Insurance — Rate Filing Guidelines.
• OSFI (2017). Guideline E-23 — Model Risk Management.
• AMF (2023). Lignes directrices — Tarification de l'assurance automobile.

---

Reda Hakkani — PhD Candidate, Applied Mathematics | Montréal, QC
Available for actuarial and quantitative risk roles — hakkanireda@hotmail.com