Logo Lanfrica

Candra0x6/Shira

Domaine:

socioeconomic

Type de record:

model
Créateur:
Can
Hôte:
# 🕌 Shira: Shariah Compliance Prediction Model > **Real-time Islamic finance compliance screening for 605 Indonesian companies using machine learning** ## 📊 Executive Summary This project delivers a **production-ready machine learning system** that predicts Shariah (Islamic finance) compliance for Indonesian Stock Exchange (IDX) listed companies using **real financial data (2020-2023)** from 605 companies. ### ✨ Key Achievement Transformed the project from synthetic data to **real Indonesian financial statements** (89,243 records) and trained an XGBoost model achieving: | Metric | Value | Status | |--------|-------|--------| | **Test Accuracy** | 92.00% | ✅ Validated | | **ROC-AUC** | 92.83% | ✅ High Confidence | | **Training Companies** | 496 | ✅ Real IDX Data | | **Features Engineered** | 12 Shariah-compliant ratios | ✅ Domain-aligned | | **Compliance Rate** | 22.2% compliant | ✅ Realistic for Indonesia | | **Model Size** | 4.2 MB | ✅ Lightweight | --- ## 🎯 Business Problem & Solution ### The Problem - Indonesian Islamic finance requires **Shariah compliance screening** for all investments - Manual audits are **expensive, slow, and subjective** - Regulators need **consistent, auditable, transparent** decision-making - Current rules-based approaches miss **nuanced financial patterns** ### The Solution **ML-powered compliance screening** that: - ✅ Analyzes 12 financial ratios per company - ✅ Applies OJK/DSN-MUI Islamic finance rules - ✅ Provides confidence scores and explanations - ✅ Processes 605 companies in 60%) OR (Interest Income Ratio > 10%) OR ... → Non-Compliant (FAIL) ELSE IF (All Financial Rules Pass) → Compliant (PASS) ELSE → Borderline (Manual Review) ``` --- ## 📈 Results & Insights ### Compliance Distribution ``` Total Companies Analyzed: 495 ✅ Shariah Compliant: 110 companies (22.2%) ❌ Non-Compliant: 385 companies (77.8%) Compliance Breakdown: ├── By Sector Compliance: │ ├── Oil & Gas: 1/5 (20.0%) │ ├── Other: …

Languages