Provenance-and machine learning-based recommendation of parameter values in scientific workflows.

Area of Science:

• Computational science
• Bioinformatics
• Data science

Background:

• Scientific Workflows (SWfs) are essential tools for modern scientific experimentation.
• Managing SWfs involves complex tools for composition, execution, and data handling.
• Incorrect parameter configuration in SWfs can lead to execution failures and wasted resources.

Purpose of the Study:

• To introduce FReeP (Feature Recommender from Preferences), a novel method for recommending parameter values in SWfs.
• To leverage Machine Learning, specifically Preference Learning, for automated parameter suggestion.
• To reduce the risk of workflow failures and optimize resource utilization.

Main Methods:

• FReeP employs three Machine Learning algorithms based on Preference Learning.
• Two algorithms focus on recommending values for individual parameters.
• A third algorithm is designed to recommend values for multiple parameters simultaneously.

Main Results:

• Experimental results using provenance data from two widely used SWfs demonstrate FReeP's effectiveness in recommending single parameter values.
• The study indicates FReeP's potential for recommending values for multiple parameters in SWfs.
• FReeP successfully utilizes past user preferences to guide recommendations.

Conclusions:

• FReeP offers a promising solution for automating parameter configuration in Scientific Workflows.
• The method has shown utility in improving the reliability and efficiency of SWf execution.
• Further development could enhance its capability for complex, multi-parameter recommendations.