kirikCS

kirikCS / data-fusion-guardian-ICEQ

Public

Репозиторий SOTA паблик-решения команды ICЕQ

12
3
100% credibility
Found Mar 29, 2026 at 12 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Jupyter Notebook
AI Summary

A complete solution with scripts, models, and features for classifying unconfirmed fraudulent banking operations in a competition dataset, achieving the top public leaderboard score of 0.1445.

How It Works

1
🔍 Discover the Fraud Buster

You find this handy tool on a competition page that helps spot sneaky unconfirmed bank transactions in huge lists of customer activity.

2
📥 Grab the Files

Download the folder with ready-made results and simple guides to get started right away.

3
Pick Your Speed
⏰
Instant Win

Use the pre-made results file to jump straight to uploading.

🚀
Quick Refresh

Run a short process on your computer to update with perfect blends.

🔮
Full Power

Let it crunch all data through steps to create top-notch predictions.

4
💫 Watch It Work

Your computer hums along, turning raw transaction stories into smart fraud alerts with special customer patterns.

5
📊 Get Your Results

Out pops a file ready with fraud scores for every transaction, tuned for top accuracy.

🏆 Score Big

Upload your file and celebrate a leaderboard score of 0.1445, proving you've caught the fraudsters!

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 12 to 12 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is data-fusion-guardian-ICEQ?

This repo delivers a battle-tested pipeline for fraud detection in banking transactions, classifying unconfirmed operations (RED fraud class) on 200M+ events from 100K clients using Python scripts optimized for GPU. It generates hundreds of features like device fingerprints, velocity checks, and behavioral drifts, then trains CatBoost ensembles boosted by self-supervised client embeddings from GRU sequences. Users get a ready submission.csv hitting 0.1445 PR-AUC on the public leaderboard, plus quick-start scripts for full retraining in under an hour.

Why is it gaining traction?

As a SOTA AI GitHub solution for data fusion tasks like semantic segmentation or object detection analogs in tabular fraud, it stands out with fraud-type-specific features (VoIP bursts, impossible travel) and clever blending like rank-averaged logits with feedback injection. Developers grab it for the transparent experiment logs—what works (+0.015 from feedback), what flops (GRU direct prediction)—saving weeks on feature ideation. The CoLES embeddings add behavioral client profiles without labels, hooking Kaggle pros chasing leaderboard edges.

Who should use this?

Fraud ML engineers at banks tuning PR-AUC on imbalanced transaction logs, Kaggle tabular competitors baseline-ing CatBoost setups, or fintech teams prototyping device-sharing detectors. Ideal if you're handling timestamped events with sparse labels and need GPU-accelerated feature gen.

Verdict

Fork it for a strong fraud baseline—detailed README and ready models make it plug-and-play despite 12 stars and 1.0% credibility score. Maturity is early (low tests, competition-specific), but experiment writeups make it a learning goldmine over generic boosters.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.