After having used KNIME for a while and helped colleagues applying it for their data analysis tasks, I found that one of the barriers for users that are new to data mining is the choice of algorithms. While concept like CRISP-DM apparently leave this one out in their modeling steps, the e-LICO project seems to have found a way to use meta-learning on former (successful) mining workflows to predict (and evaluate) the right choice of algorithm.
Not only that, they seem to have developed an add-in for rapidminer to do that.
While KNIME already supports the recommendation feature coined "Workflow Coach", I did not yet come across a similarly sophisticated support for the choice of algorithm within KNIME. Did I miss something or is it maybe just not that interesting for the DM-savvy KNIME power user?
Thank you for your feedback.
At the moment KNIME Analytics Platform does not provide a feature that support for the choice of algorithms.
May I ask you what would be your expectation about a recommendation engine that support for the choice of the algorithm to use in KNIME Analytics Platform?
I've been working with KNIME for a while and love to brag about the user-oriented workflow/pipeline interface that reduces the need to code. However, the user orientation ends there as e.g. business analysts might not be confident in choosing the adequate algorithm to serve their need. As a result I've lots of nice workflows with doubtful mining algorithm choices.
It would be great if - depending on input data and analysis task in question - appropriate algorithms could be recommended by KNIME, e.g. based on empirical results from all KNIME installations.