Model Performance Benchmarking & Visualization

[hoangsonww]

Summary
Add a benchmarking module that evaluates all classifiers (vehicle, face, mood, speech, sentiment, etc.) on their respective datasets and produces performance metrics + visualizations (accuracy, precision, recall, F1-score, confusion matrix).

---

Motivation / Why

• Currently, classifiers can be run individually, but users don’t have a clear way to measure or compare their performance.

• A standardized benchmarking module would:

  • Help developers understand strengths/weaknesses of each model.
  • Allow reproducible experiments when retraining or fine-tuning.
  • Enable easy reporting for academic or demo purposes.

---

Proposed Solution

• Evaluation Script:

  • Create benchmark.py that loads test datasets for each classifier.
  • Compute metrics: accuracy, precision, recall, F1-score, confusion matrix.

• Visualization:

  • Use Matplotlib/Seaborn to plot confusion matrices and ROC curves.
  • Save results in a /results directory as images and CSV/JSON logs.

• Integration:

  • Extend main.py to include a "Run Benchmark" option.
  • Optionally expose results via Flask app for interactive browsing.

---

Alternatives Considered

• Manually testing each classifier with sample inputs (time-consuming and inconsistent).
• Relying on logs only, but this doesn’t provide visual insights.

---

Additional Context

• Could integrate with MLflow or Weights & Biases (already listed in the tech stack) for experiment tracking.
• Makes the project more attractive for research, education, and reproducibility.

---

✅ Acceptance Criteria

• Implement benchmark.py script for running tests across classifiers.
• Generate confusion matrices & metric reports for each classifier.
• Save/export results to /results directory in CSV/JSON format.
• Optional: Flask UI extension to display benchmark results interactively.
• Update documentation (README.md) with benchmarking usage guide.

---