README.md

metadata

title: Hybrid AI Fraud Detection XAI Dashboard
emoji: 🛡️
colorFrom: blue
colorTo: green
sdk: streamlit
sdk_version: 1.28.0
app_file: app.py
pinned: true
license: mit
tags:
  - fraud-detection
  - explainable-ai
  - lightgbm
  - lstm
  - graph-neural-network
  - shap
  - ieee-cis

Hybrid AI Architectures for Proactive Fraud Detection

An XAI-Driven Optimization Framework

Auteur: Latif SINARE Programme: Master MAII - FSTH, Universite Abdelmalek Essaadi

Objectifs quantitatifs

Recall >= 90% - Precision >= 10% - AUPRC > 0,85 - FPR < 0,01%

Pipelines

Phase	Dataset	Modeles	Optimiseur
1	Kaggle CC	LightGBM	CMA-ES
2	PaySim	LSTM + RGAT + Stacking	PSO Async
3	IEEE-CIS	LSTM + RGAT + Stacking	PSO Async
4	AMLSim	RGAT	-

Pages du dashboard

1. Pipeline 1 - Kaggle CC: analyse interactive de transaction + SHAP
2. Pipeline 2 - PaySim: resultats Stacking + analyse heuristique
3. Pipeline 3 - IEEE-CIS: resultats complets + PSI drift monitoring
4. Performances: comparaison des 3 phases

References

• Lundberg and Lee (2017) - SHAP
• Lopez-Rojas et al. (2016) - PaySim
• Wang et al. (2019) - GNN Fraud Detection