XGPrice

Technical-first machine learning project for stock movement modeling with a focus on XGBoost pipelines and notebook-driven experimentation.

Project Scope

• Forecasting / classifying price behavior for ASELSAN-based market data.
• Feature engineering and preprocessing for both classification and regression tracks.
• Experimentation in notebooks and script exports.
• Model artifact persistence and result specs under models/.

Current Stack

• Python 3.10+
• XGBoost (XGBClassifier, XGBRegressor)
• scikit-learn (metrics, split logic, utilities)
• pandas / NumPy
• matplotlib / seaborn
• Optuna (hyperparameter search)
• SHAP (model explainability experiments)

Repository Layout

.
├─ data/
│  ├─ gfinance/
│  ├─ tradingview/
│  └─ yfinance/
├─ export/
│  ├─ predict.py
│  ├─ xgb_classification.py
│  └─ xgb_regression.py
├─ models/
│  ├─ *.model
│  └─ *_model_specs.txt
└─ src/
   ├─ DL/
   ├─ preprocess/
   ├─ XGB/
   └─ other/

Workflow Summary

1. Prepare data in preprocessing modules / notebooks.
2. Train with XGBoost workflows (classification or regression).
3. Evaluate with fold-based metrics and confusion/error analysis.
4. Save model checkpoints and experiment specifications to models/.
5. Run prediction scripts for downstream inspection/backtesting.

Reproducibility Quick Start

1) Environment

Install dependencies:

pip install -r requirements.txt

2) Preferred execution

• Open and run notebooks under src/XGB/ and src/DL/ for iterative experimentation.
• Use scripts in export/ for exported notebook workflows.

3) Data and model artifacts

• Input and processed datasets are kept in data/.
• Checkpoints and spec reports are kept in models/.

Baseline Snapshot (from committed specs)

From models/multi_mk1_xgboost_model_specs.txt:

• Multi-class AUC: 0.6247
• Accuracy: 0.4261
• Backtest total balance: $286,814.59
• Backtest stock difference: 142.51%

These values are historical run outputs and should be interpreted as experiment-specific, not guaranteed future performance.

Engineering Notes

• This repository intentionally tracks:
  • .ipynb notebooks
  • dataset files under data/
  • model checkpoints/specs under models/
• Temporary scratch file src/XGB/temp.txt is excluded from version control.

Disclaimer

This is a research/engineering project. It is not financial advice.