---
library_name: lightgbm
tags:
- cryptocurrency
- blockchain
- fraud-detection
- illicit-activity-detection
- graph-analysis
- bitcoin
- ethereum
- solana
- cybersecurity
pipeline_tag: tabular-classification
---

# SIH26183 Universal Cross-Chain Risk Model

A structure-based global LightGBM model for cryptocurrency
transaction-flow risk triage.

## Supported chains

- Bitcoin
- Ethereum
- Solana

## Model architecture

Normalized blockchain transactions
→ directed flow graph
→ 12 universal structural features
→ StandardScaler
→ LightGBM
→ risk probability

## Universal features

[
  "tx_count",
  "unique_counterparties",
  "in_tx_count",
  "out_tx_count",
  "in_degree",
  "out_degree",
  "total_degree",
  "degree_ratio",
  "is_source",
  "is_sink",
  "mean_successor_out_degree",
  "mean_predecessor_in_degree",
  "pagerank",
  "active_span",
  "mean_event_time",
  "event_time_std",
  "recent_activity_age",
  "log_in_volume",
  "log_out_volume",
  "log_total_volume",
  "in_out_value_ratio",
  "inflow_concentration",
  "outflow_concentration"
]

## Threshold

0.45545454545454545

## Important limitation

This model provides an analytical triage signal.

It does NOT independently establish:

- criminal liability
- wallet ownership
- VASP attribution
- intent
- legal guilt

Blockchain evidence and investigator review remain authoritative.

## Input

The inference service expects normalized transaction
relationships following the CommonTransaction structure.

## Output

The service returns:

- ML probability
- threshold decision
- risk level
- risk points
- graph statistics
- per-node scores

## Intended use

Investigation-support and research purposes.