How alokyadav310703/ieee-cis-transaction-fraud-detection-system works
This repository is a fraud-detection ML project with a Streamlit demo. The documented workflow takes IEEE-CIS transaction and identity data through preprocessing, feature engineering, model training and evaluation; the demo then transforms sample rows with saved preprocessing artifacts, scores them with LightGBM, and presents predictions and evaluation. The training notebook is documented but unsampled, so its detailed internal calls are not mapped. Relationships below rely on README statements or the shown app and pipeline excerpts.
Main entry points: 02_feature_engineering_&_model_training.ipynb Training pipeline · streamlit_app.py Streamlit demo · 04_save_pipeline_artifacts.py Artifact export · pipeline_artifacts.pkl Pipeline artifacts
Components 18
Data
- IEEE-CIS data
- Holdout sample
data/sample_1000.csv
Model development
- Exploratory analysis
notebooks/01_eda.ipynb - Training pipeline
notebooks/02_feature_engineering_&_model_training.ipynb - Shared preprocessing
fraud_pipeline.py - Feature engineering
fraud_pipeline.py - SMOTE balancing
notebooks/02_feature_engineering_&_model_training.ipynb - Model training
notebooks/02_feature_engineering_&_model_training.ipynb - Rank ensemble
notebooks/02_feature_engineering_&_model_training.ipynb - Artifact export
notebooks/04_save_pipeline_artifacts.py
Demo scoring
- Pipeline artifacts
models/pipeline_artifacts.pkl - LightGBM model
models/lgb_gbdt.txt - Streamlit demo
streamlit_app.py - Inference transform
fraud_pipeline.py - Transaction scoring
streamlit_app.py
Results
- Model evaluation
notebooks/02_feature_engineering_&_model_training.ipynb - Metrics and dashboard
streamlit_app.py
Other components
- Analyst
Connections 22
- Analyst → Exploratory analysis: explores data
README.md - Analyst → Training pipeline: runs training
README.md - IEEE-CIS data → Training pipeline: supplies data
README.md - IEEE-CIS data → Exploratory analysis: supplies data
README.md - Training pipeline → Shared preprocessing: fits preprocessing
fraud_pipeline.py - Training pipeline → Feature engineering: engineers features
README.md - Training pipeline → SMOTE balancing: applies SMOTE
README.md - Training pipeline → Model training: trains models
README.md - Training pipeline → Rank ensemble: combines predictions
README.md - Training pipeline → Model evaluation: evaluates models
README.md - Analyst → Artifact export: exports artifacts
notebooks/04_save_pipeline_artifacts.py - Artifact export → Shared preprocessing: calls fit
notebooks/04_save_pipeline_artifacts.py - Artifact export → Pipeline artifacts: writes artifacts
notebooks/04_save_pipeline_artifacts.py - Model training → LightGBM model: saves model
README.md - Analyst → Streamlit demo: starts demo
streamlit_app.py - Holdout sample → Streamlit demo: loads sample
streamlit_app.py - Pipeline artifacts → Streamlit demo: loads artifacts
streamlit_app.py - LightGBM model → Streamlit demo: loads model
streamlit_app.py - Streamlit demo → Inference transform: calls transform
fraud_pipeline.py - Inference transform → Transaction scoring: returns features
fraud_pipeline.py - Streamlit demo → Transaction scoring: simulates transactions
streamlit_app.py - Transaction scoring → Metrics and dashboard: shows results
streamlit_app.py