Fraud Detection System

Predictive analysis of auto claim irregularities
Single Claim Analysis
Batch Upload Analysis
Risk drivers are generated using a ExtraTrees uncalibrated reference model for consistency and interpretability.
Prediction scores are produced by the selected model above.
Scenario determines scoring behavior (e.g. Model Calibration), while Operating Mode sets the review posture (e.g. Triage Thresholds).
01 / Incident & Injury
12:00
Afternoon
02 / Claim Amount
$58,000
Total amount claimed before internal allocation.
Claim Amount Breakdown
Estimated distribution of the total claim amount (percentages are relative and do not need to sum perfectly)
10%
Medical and bodily injury related costs
10%
Vehicle and physical property damage
80%
Remaining claim components (legal, admin, other)
03 / Policy & Tenure
$1,200
200
04 / Vehicle Context
10 yrs
05 / Administrative
06 / Edge Variables
38 yrs
Financial Capital Activity (Optional)
Summary of observed capital inflows and outflows around the incident period.
$0
$0

Data Requirements

1. Flexibility Column order does not matter. The system matches data by Header Name (e.g., age), so you can arrange columns however you like.
2. Robustness (Missing Data) Missing columns will not crash the model. We apply smart defaults for any missing field (e.g., age=38, hour=12) to ensure you always get a prediction.
3. Strict Rules
  • Headers: Must match the template names exactly (case-sensitive).
  • Claim ID: Optional but highly recommended. We generate SYS_GEN_XXXX IDs if omitted.
Identifiers
claim_id
HIGHLY RECOMMENDED. Unique ID for the claim. Providing this allows you to tie the results/decisions back to your internal batches.
String / Integer
If not provided, the system will generate its own IDs, but this may make matching results difficult.
Financial & Policy
policy_annual_premium
Annual cost of the insurance policy.
Float • e.g. 1200.50
total_claim_amount
Total value of the claim requested.
Float • e.g. 50000
umbrella_limit
Umbrella policy coverage limit.
Integer • 0, 1000000, 5000000, 10000000
capital-gains
Reported capital gains valid for the period. Note the hyphen.
Float • >= 0
capital-loss
Reported capital loss. Provide as positive number or negative (logic handles both).
Float
Incident Details
collision_type
Type of impact.
Options: Front Collision, Rear Collision, Side Collision, ?
incident_severity
Assessed damage severity.
Options: Major Damage, Minor Damage, Total Loss, Trivial Damage
incident_hour_of_the_day
Hour when incident occurred (24h format).
Integer • 0-23
authorities_contacted
Agency contacted at scene.
Options: Police, Fire, Ambulance, Other, None
police_report_available
Whether a report exists.
Options: YES, NO, ?
number_of_vehicles_involved
Integer • >= 1
bodily_injuries
Integer • >= 0
Customer History
months_as_customer
Total tenure in months.
days_since_bind
Days since the current policy was bound/started.
age
Insured's age in years.
vehicle_age
Vehicle age in years.
Breakdown Shares (Optional)
injury_share
Percentage of claim for injury. (0-100)
property_share
Percentage of claim for property. (0-100)