Built an automated benchmarking pipeline comparing Logistic Regression, Random Forest, XGBoost, and Neural Networks using Optuna hyperparameter tuning.
In real-world ML systems, model selection is not just about accuracy but also latency and scalability.
This project demonstrates how simpler models can outperform complex ones while being significantly faster, which is critical for production environments.
- F1-score: 0.986
- PR-AUC: 0.997
- Latency: ~2.7 ms
A key finding was that simpler models (Logistic Regression) outperformed more complex models like XGBoost and Neural Networks while achieving 2–4× lower inference latency. This highlights the importance of balancing performance with efficiency in production ML systems.
Python, Scikit-learn, XGBoost, Optuna, NumPy, Pandas, Matplotlib
pip install -r requirements.txt python benchmark.py