CyberShield 2026 · Bank of India × IIT Hyderabad
Mule accounts, intercepted
before the money moves.
A leakage-audited ML ensemble screens every account against 2,116 behavioural features, explains each verdict with SHAP, and files regulator-ready STRs — built on the real Bank of India dataset.
ROC-AUC
0.9913
Avg Precision
0.9365
Recall @ F1-opt
91.4%
Accounts screened
9,082
Surveillance Feed
Live · sync 6s| Time | Account | Channel | Amount | Risk |
|---|---|---|---|---|
| 14:02:11 | ACC-3187 | IMPS OUT | ₹9,40,000 | Critical83.0 |
| 14:01:47 | ACC-1996 | NEFT IN | ₹12,00,000 | Critical94.0 |
| 14:01:22 | ACC-7742 | UPI IN | ₹24,500 | Medium29.0 |
| 14:00:58 | ACC-2380 | RTGS OUT | ₹6,10,000 | High61.0 |
| 14:00:31 | ACC-5521 | IMPS IN | ₹8,200 | Low8.0 |
| 14:00:04 | ACC-9044 | NEFT OUT | ₹45,000 | Low12.0 |
Platform Modules
09 modulesAccount Screening
Per-account risk verdict in under a second, with per-model scores.
Bulk CSV Screening
Upload a full dataset — every account scored in one pass.
Model Performance
ROC / PR curves and confusion matrix from out-of-fold predictions.
SHAP Explainability
Feature-level reasons for every flag. Defensible in audit.
Fund-Flow Graph
Money movement across accounts, branches and banks.
STR Filing
One-click Suspicious Transaction Report. PMLA Rule 8 format.
Anomaly Lab
Unsupervised contrastive layer outside the labelled set.
Live Feeds & Compliance
Streaming alerts with RBI / PMLA alignment tracking.
Cost-Benefit & Case Story
714% ROI economics; GenAI officer-ready case narrative.