AutoML vs Hyperparameter optimization: A case study in software defect prediction

Take home:

HPT xgboost always achieve the best performance. However, for the small dataset (i.e., camel 1.2), it takes a longer time to achieve 99% close-to-the-highest performance.
AML and HPT can both achieve 95% close-to-the-highest AUC_weighted within 13 iterations, So we don’t have to hyper-parameter tune/deploy AML for a lot of iterations.
Possible advice for the practitioner would be to run the AML for 20 iterations and Hyperparameter tune the suggested model for the best performance.
The interpretation varies between the models tuned by AML and HPT. Therefore, we recommend using the interpretation of the model with the best fit.

Name		Name	Last commit message	Last commit date
Latest commit History 12 Commits
data		data
estimator		estimator
hyper_model		hyper_model
test		test
AutoML.ipynb		AutoML.ipynb
Compare_performance.ipynb		Compare_performance.ipynb
Estimator_test.ipynb		Estimator_test.ipynb
Hyperparameter_tuning.ipynb		Hyperparameter_tuning.ipynb
Iteration_analysis_A.ipynb		Iteration_analysis_A.ipynb
Iteration_analysis_H.ipynb		Iteration_analysis_H.ipynb
README.md		README.md
Top5_loaded_feature_importance.ipynb		Top5_loaded_feature_importance.ipynb
Visualize_A.ipynb		Visualize_A.ipynb
Visualize_H.ipynb		Visualize_H.ipynb