Student Cooperative Micro-Credit Scoring & Fraud Alert System
Digital transformation in student cooperatives and community micro-finance requires automated risk assessment for small working capital loans. Construct a lightweight credit scoring agent using XGBoost and Polars to evaluate loan application risks. Achieve high default prediction accuracy while ensuring zero demographic bias.
What you are building
Build an automated micro-credit risk evaluation and fraud alert pipeline using Polars and XGBoost.
How work is evaluated
Evaluates default prediction ROC-AUC score and fair-lending threshold compliance.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
risk_threshold_check
Verifies that high default risk results in immediate REJECT decision.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
roc_auc
Area under ROC curve for default prediction model. • target: 0.88 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Ingest high-throughput transaction telemetry using Polars DataFrames
Train gradient boosted decision trees with XGBoost for default probability estimation
Enforce fairness constraints to prevent biased credit decisioning
Expose fast evaluation routines for embedded micro-finance mobile apps
Resources
Synthetic micro-finance loan applicant repayment dataset.
Challenge setup
Evaluates default prediction ROC-AUC score and fair-lending threshold compliance.
Challenge input
JSON applicant financial profile
Polars
High performance Rust-backed DataFrame library.
Evaluated output
JSON decision and risk score
- Verifies that high default risk results in immediate REJECT decision.
- Area under ROC curve for default prediction model. • target: 0.88 • range: 0-1
- Benchmark: MicroCredit-FairRisk-Bench
- Roc Auc target: 0.88
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- Capably · Required
- AutoML (H2O) · Optional
- MemGPT · Optional
Evaluation contract
- risk_threshold_check · Weight 1
- roc_auc · Weight 1
Metrics
- roc_auc · Target 0.88
Benchmarks
- MicroCredit-FairRisk-Bench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON applicant financial profile
Output: JSON decision and risk score
Execution harness: Python
Evaluation tools
- Polars · DataFrame Engine
Recipe freeze marker
No recipe freeze time is recorded.