End user friendly data mining with decision trees: a reality or a wish?
Petra Povalej, Peter Kokol · 2007
Abstract:- The main focus of data mining is to present hidden knowledge located in large amount of data in human understandable form. Therefore the knowledge representation has to be simple and easy to interpret, possibly without the computer. Decision trees are one of the most transparent methods often used in data mining, but can we make them user friendly? In the process of decision tree induction a lot of input parameters have to be fine-tuned in order to obtain good results. To brain the right combination of input parameters for a specific problem is a hard task usually performed by data mining expert. So, to make decision tree based data mining end user friendly we explored various alternatives of decision tree induction, concentrating on purity measures. We introduced new hybrid purity measures and tested their adequacy on real world databases. Additionally we constructed a meta decision tree to determine the best combination of input parameters. Key-Words:- knowledge discovery, decision trees, purity measures, hybrid approach 1