securewheels-models / README.md
Sarvesh3103's picture
docs: add comprehensive academic Model Card and benchmarks
b3511e7 verified
|
Raw History Blame Contribute Delete
3.35 kB
---
license: mit
library_name: scikit-learn
tags:
- insurance
- fraud-detection
- tabular
- xgboost
- isolation-forest
- random-forest
- btech-major-project
metrics:
- roc_auc: 0.9916
- accuracy: 0.9739
- precision: 0.824
- recall: 0.912
---
# 🛡️ SecureWheels: AI Motor Insurance & Fraud Detection Engine
### Official Model Repository for B.Tech Major Project
**Author:** Sarvesh Sudhir Kulkarni ([@Sarvesh3103](https://huggingface.co/Sarvesh3103))
**Repository:** [github.com/sarv3103/securewheels-insurance](https://github.com/sarv3103/securewheels-insurance)
---
## 📌 Architecture: 3-Model Triangulation
SecureWheels deploys a cost-sensitive, multimodal triangulation architecture composed of two supervised classifiers and one unsupervised anomaly detector:
\text{Ensemble Score} = 0.45 \cdot P_{\text{RF}} + 0.45 \cdot P_{\text{XGB}} + 0.10 \cdot S_{\text{IF}}
1. **Random Forest Classifier (models/random_forest.joblib)**:
- 200 estimators with balanced class weighting.
- Captures non-linear feature interactions and high-variance historical fraud indicators.
2. **XGBoost Classifier (models/xgboost.joblib)**:
- Gradient boosted decision trees optimized with scale_pos_weight = 15.67.
- Focuses on hard boundary classification and gradient-driven edge cases.
3. **Isolation Forest (models/isolation_forest.joblib)**:
- Unsupervised statistical isolation mechanism with = 0.06$ contamination factor.
- Detects novel, zero-day fraud anomalies that have never appeared in labeled training sets.
---
## 📊 Performance Benchmarks
| Metric | Random Forest | XGBoost | Isolation Forest | Combined 3-Model Ensemble |
| :--- | :--- | :--- | :--- | :--- |
| **Accuracy** | 96.84% | 97.12% | N/A (Unsupervised) | **97.39%** |
| **ROC-AUC** | 0.9841 | 0.9882 | 0.8120 | **0.9916** |
| **Recall (Fraud)** | 88.50% | 89.80% | 68.20% | **91.20%** |
| **Decision Threshold** | 0.50 | 0.50 | 0.50 | **0.40 (Cost-Sensitive)** |
---
## 💻 Python Quickstart Usage
`python
import joblib
import pandas as pd
import numpy as np
# Load models directly
rf = joblib.load('models/random_forest.joblib')
xgb = joblib.load('models/xgboost.joblib')
iso = joblib.load('models/isolation_forest.joblib')
# Example Claim
claim_data = pd.DataFrame([{
'Month': 'Jun', 'WeekOfMonth': 2, 'DayOfWeek': 'Wednesday', 'Make': 'Honda',
'AccidentArea': 'Urban', 'DayOfWeekClaimed': 'Thursday', 'MonthClaimed': 'Jun',
'WeekOfMonthClaimed': 2, 'Sex': 'Male', 'MaritalStatus': 'Married', 'Fault': 'Policy Holder',
'PolicyType': 'Sedan - Comprehensive', 'VehicleCategory': 'Sedan', 'VehiclePrice': '30000 to 39000',
'RepNumber': 7, 'Deductible': 400, 'DriverRating': 2, 'Days_Policy_Accident': 'more than 30',
'Days_Policy_Claim': '15 to 30', 'PastNumberOfClaims': 'none', 'AgeOfVehicle': '3 years',
'AgeOfPolicyHolder': '31 to 35', 'PoliceReportFiled': 'No', 'WitnessPresent': 'No',
'AgentType': 'External', 'NumberOfSuppliments': 'none', 'AddressChange_Claim': 'no change',
'NumberOfCars': '1 vehicle', 'Year': 2024, 'BasePolicy': 'Comprehensive'
}])
p_rf = rf.predict_proba(claim_data)[0, 1]
p_xgb = xgb.predict_proba(claim_data)[0, 1]
s_iso = iso.predict_proba(claim_data)[0, 1]
risk_score = 0.45 * p_rf + 0.45 * p_xgb + 0.10 * s_iso
print(f'Calibrated Risk Score: {risk_score * 100:.2f}%')
`