Multiple Decision Tree configurations were evaluated on the Palmer Penguins classification task.
- The baseline Decision Tree recorded a Precision and ROC-AUC of 0.97.
- EXP-02 improved the recorded Precision and ROC-AUC to 0.98 after modifying the model configuration and preprocessing.
- EXP-03 produced the highest recorded Precision of 0.99 and ROC-AUC of 0.99.
- The pruned Decision Tree in EXP-04 retained a ROC-AUC of 0.99 while recording a Precision of 0.98.
- Recording each experiment separately made it easier to compare changes in model configuration, preprocessing, and evaluation metrics.
The project demonstrates the importance of maintaining a structured history of machine learning experiments.
Instead of modifying a model without recording previous results, each experiment captures:
- Model configuration
- Hyperparameters
- Preprocessing choices
- Feature-selection choices
- Evaluation metrics
Git-based version control provides an additional history of changes made during experimentation.
Among the recorded experiments, EXP-03 achieved the strongest overall result, with both Precision and ROC-AUC reaching 0.99.
The project provided a foundational introduction to experiment tracking before moving to automated tools such as MLflow in more advanced MLOps workflows.