File size: 5,293 Bytes
9bbbb5b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
"""
Synthetic Training Data Generator for Insurance Claims Decision Support
========================================================================

GOVERNANCE CONSTRAINTS:
- Generates data ONLY for features defined in decision_spec.yaml
- Uses FROZEN decision boundaries to assign labels
- Synthetic data for demonstration purposes only
- No real customer data used

Purpose: Create training dataset with proper input features
"""

import pandas as pd
import numpy as np
import random
from datetime import datetime

# FROZEN DECISION BOUNDARIES - DO NOT MODIFY
DECISION_BOUNDARIES = {
    'damage_thresholds': {
        'low': 5000,
        'medium': 15000,
        'high': 50000
    },
    'risk_weights': {
        'low': 1.0,
        'medium': 1.5,
        'high': 2.0
    },
    'injury_multiplier': 1.8,
    'severity_thresholds': {
        'low': 5,
        'medium': 15
    }
}

def calculate_severity_score(claim_type, damage_amount, injury_involved, risk_factor):
    """
    Calculate severity score using FROZEN decision boundaries.
    This replicates the logic from decision_spec.yaml.
    """
    # Base score from damage amount
    if damage_amount < DECISION_BOUNDARIES['damage_thresholds']['low']:
        damage_score = 2
    elif damage_amount < DECISION_BOUNDARIES['damage_thresholds']['medium']:
        damage_score = 5
    elif damage_amount < DECISION_BOUNDARIES['damage_thresholds']['high']:
        damage_score = 10
    else:
        damage_score = 20
    
    # Apply risk weight
    risk_weight = DECISION_BOUNDARIES['risk_weights'][risk_factor]
    score = damage_score * risk_weight
    
    # Apply injury multiplier
    if injury_involved:
        score *= DECISION_BOUNDARIES['injury_multiplier']
    
    # Determine severity level
    if score < DECISION_BOUNDARIES['severity_thresholds']['low']:
        return 'low'
    elif score < DECISION_BOUNDARIES['severity_thresholds']['medium']:
        return 'medium'
    else:
        return 'high'

def generate_synthetic_dataset(n_samples=1000, random_seed=42):
    """
    Generate synthetic training data based on decision_spec.yaml.
    
    Args:
        n_samples: Number of samples to generate
        random_seed: Random seed for reproducibility
    
    Returns:
        DataFrame with input features and target labels
    """
    random.seed(random_seed)
    np.random.seed(random_seed)
    
    print("=" * 70)
    print("GENERATING SYNTHETIC TRAINING DATA")
    print("=" * 70)
    print(f"Samples to generate: {n_samples}")
    print(f"Random seed: {random_seed}")
    print(f"\nFeatures (from decision_spec.yaml):")
    print("  - claim_type: categorical (Auto, Property, Health, Liability)")
    print("  - damage_amount: numeric (USD)")
    print("  - injury_involved: boolean")
    print("  - risk_factor: categorical (low, medium, high)")
    print(f"\nTarget: severity (low, medium, high)")
    print(f"Calculation: Using FROZEN decision boundaries")
    
    data = []
    
    for i in range(n_samples):
        # Generate random input features
        claim_type = random.choice(['Auto', 'Property', 'Health', 'Liability'])
        
        # Generate damage amount with realistic distribution
        # Log-normal distribution for realistic claim amounts
        damage_amount = np.random.lognormal(mean=9, sigma=1.2)
        damage_amount = round(min(damage_amount, 200000), 2)  # Cap at $200k
        
        # Injury more likely for Auto and Liability claims
        if claim_type in ['Auto', 'Liability']:
            injury_involved = random.choices([True, False], weights=[0.3, 0.7])[0]
        else:
            injury_involved = random.choices([True, False], weights=[0.1, 0.9])[0]
        
        # Risk factor distribution
        risk_factor = random.choices(
            ['low', 'medium', 'high'],
            weights=[0.5, 0.35, 0.15]
        )[0]
        
        # Calculate severity using FROZEN boundaries
        severity = calculate_severity_score(
            claim_type, damage_amount, injury_involved, risk_factor
        )
        
        data.append({
            'claim_type': claim_type,
            'damage_amount': damage_amount,
            'injury_involved': injury_involved,
            'risk_factor': risk_factor,
            'severity': severity
        })
    
    df = pd.DataFrame(data)
    
    print(f"\n{'='*70}")
    print("DATASET GENERATION COMPLETE")
    print(f"{'='*70}")
    print(f"Total samples: {len(df)}")
    print(f"\nFeature summary:")
    print(df.describe(include='all'))
    print(f"\nTarget distribution:")
    print(df['severity'].value_counts())
    print(f"\nSample rows:")
    print(df.head(10))
    
    return df

if __name__ == "__main__":
    # Generate dataset
    df = generate_synthetic_dataset(n_samples=1000)
    
    # Save to CSV
    output_file = 'synthetic_training_data.csv'
    df.to_csv(output_file, index=False)
    print(f"\n{'='*70}")
    print(f"Dataset saved to: {output_file}")
    print(f"{'='*70}")
    print("\nGOVERNANCE STATUS: ✓ COMPLIANT")
    print("  - Uses only allowed features from decision_spec.yaml")
    print("  - Applies FROZEN decision boundaries")
    print("  - Synthetic data (no real customer information)")
    print("  - Suitable for demonstration/training purposes only")