app.py

import streamlit as st
import json, os

st.set_page_config(
    page_title="Hybrid AI Fraud Detection",
    page_icon="shield",
    layout="wide",
    initial_sidebar_state="expanded"
)

@st.cache_resource
def load_results():
    results = {}
    for name, path in [
        ("p1", "models/p1_lgbm_results.json"),
        ("p2", "models/p2_stacking_results.json"),
        ("p3", "models/p3_pipeline_results.json"),
    ]:
        if os.path.exists(path):
            with open(path) as f:
                results[name] = json.load(f)
    return results

results = load_results()

st.title("Hybrid AI Architectures for Proactive Fraud Detection")
st.subheader("An XAI-Driven Optimization Framework")
st.markdown("**Master MAII | FSTH - Universite Abdelmalek Essaadi | Latif SINARE**")
st.divider()

col1, col2, col3, col4 = st.columns(4)
with col1:
    st.metric("Recall cible", ">= 90%")
with col2:
    st.metric("FPR cible", "< 0,01%")
with col3:
    st.metric("AUPRC cible", "> 0,85")
with col4:
    st.metric("Precision min", ">= 10%")

st.divider()

col_a, col_b = st.columns(2)
with col_a:
    st.markdown("### Architecture des Pipelines")
    table_lines = [
        "| Phase | Dataset | Modeles | Optimiseur | XAI |",
        "|-------|---------|---------|------------|-----|",
        "| 1 - Baseline | Kaggle CC | LightGBM | CMA-ES | SHAP |",
        "| 2 - Prototype | PaySim | LSTM + RGAT + Stacking | PSO Async | GNNExplainer |",
        "| 3 - Production | IEEE-CIS | LSTM + RGAT + Stacking | PSO Async | SHAP + GNN |",
        "| 4 - Extension | AMLSim | RGAT | - | GNNExplainer |",
    ]
    st.markdown(chr(10).join(table_lines))

with col_b:
    st.markdown("### Fonction de Fitness Penalisee")
    st.latex(r"fitness(	heta) = AUPRC - 10\cdot\max(0, 0.90 - Recall) - 10\cdot\max(0, FPR - 0.0001)")
    st.markdown("Cette fonction garantit simultanement le Recall >= 90 pourcent et le FPR < 0,01 pourcent.")

st.divider()

if results:
    st.markdown("### Resultats disponibles")
    cols = st.columns(len(results))
    labels = {"p1": "Phase 1 - Kaggle CC", "p2": "Phase 2 - PaySim", "p3": "Phase 3 - IEEE-CIS"}
    for i, (k, res) in enumerate(results.items()):
        with cols[i]:
            th = res.get("test_holdout", res.get("fusion_test", {}))
            st.markdown(f"**{labels.get(k, k)}**")
            auprc_val = th.get("auprc", 0)
            recall_val = th.get("recall", 0)
            fpr_val = th.get("fpr", 0)
            st.metric("AUPRC", f"{auprc_val:.4f}")
            st.metric("Recall", f"{recall_val:.4f}")
            st.metric("FPR", f"{fpr_val:.6f}")
else:
    st.info("Les resultats apparaitront ici au fur et a mesure de execution des notebooks 03, 08 et 11.")

st.divider()
st.info("Utilisez la barre laterale pour naviguer entre les pipelines et analyser des transactions individuelles.")