tools/README.md

🔧 Tools Directory

Folder ini berisi model machine learning dan konfigurasi untuk anomali detection.

---

📦 Files Required

tools/
├── scaler.pkl              ← StandardScaler model (binary)
├── isolation_forest.pkl    ← Isolation Forest model (binary)
├── metadata.json           ← Configuration
├── model_train.py          ← Training script (optional)
└── simpanan_analisis.py    ← Helper functions (optional)

---

📋 File Descriptions

scaler.pkl

• Type: Binary (Pickle format)
• Purpose: Standardize features untuk ML model
• Used by: Tab 2 (Analisa Simpanan)
• Size: ~1-2 KB

isolation_forest.pkl

• Type: Binary (Pickle format)
• Purpose: Pre-trained Isolation Forest model untuk anomali detection
• Used by: Tab 2 (Analisa Simpanan)
• Features: 8 fitur (Db_Sukarela dan Cr_Sukarela aggregates)
• Size: ~10-50 KB

metadata.json

• Type: JSON (Text)
• Purpose: Konfigurasi model dan parameter
• Content:

  {
    "feature_cols": [...],           // Nama kolom untuk model
    "rolling_zscore_threshold": 1.0, // Threshold untuk Z-Score
    "rolling_window": 3              // Window size untuk rolling aggregation
  }
  

model_train.py (Optional)

Script untuk melatih ulang model jika diperlukan.

simpanan_analisis.py (Optional)

Helper functions untuk analisa simpanan.

---

⚠️ Important Notes

Binary Model Files (.pkl)

• File scaler.pkl dan isolation_forest.pkl HARUS ada untuk Tab 2 berfungsi
• Files ini di-generate saat training, bukan di-create manual
• Jangan di-edit manual - format binary

File Locations

• App mencari file di path relatif: ../tools/ (dari app folder)
• Pastikan struktur folder sesuai template

Model Updates

Jika ingin update model (retrain):

1. Jalankan python tools/model_train.py
2. File .pkl baru akan di-generate
3. Restart aplikasi

---

🔍 Troubleshooting

Error: "Model tidak ditemukan"

Solusi:
1. Pastikan scaler.pkl ada di tools/
2. Pastikan isolation_forest.pkl ada di tools/
3. Check file paths (case-sensitive di Linux/Mac)

Error: "Model load failed"

Solusi:
1. Pastikan Python version sama saat training dan running
2. Pastikan scikit-learn version sama
3. Re-train model jika masih error

---

📊 Model Info

Isolation Forest

• Algorithm: Anomaly detection via Isolation Forest
• Training data: Historical simpanan transactions
• Features: 8 aggregated metrics
  • Db_Sukarela_Total, Db_Sukarela_Avg, Db_Sukarela_Std, Db_Sukarela_Max
  • Cr_Sukarela_Total, Cr_Sukarela_Avg, Cr_Sukarela_Std, Cr_Sukarela_Max
• Output: Binary (Anomaly: -1, Normal: 1)

StandardScaler

• Purpose: Feature normalization
• Fitted on: Historical data statistics
• Usage: Pre-process features before model prediction

---

🚀 Quick Check

Verify tools setup:

cd THC_APP
python -c "import joblib; print('scaler:', joblib.load('tools/scaler.pkl')); print('model:', joblib.load('tools/isolation_forest.pkl'))"

Success jika output menunjukkan objects tanpa error.

---

Last Updated: May 2026