Parameter-Tuned Data Mining: A General Framework
Wolfgang Konen, Patrick N. Koch, Oliver Flasch, Thomas Bartz–Beielstein · 2010
Real-world data mining applications often confront us with complex and noisy data, which makes it necessary to optimize the data mining models thoroughly to achieve high-quality results. We describe in this contribution an approach to tune the parameters of the model and the feature selection conjointly. The aim is to use one framework to solve a variety of tasks. We show that tuning is of large importance for high-quality results in benchmark tasks like the Data Mining Cup: tuned models achieve rank 2 or 4 in the ranking tables, where the untuned model had rank 21 out of 67. We discuss several issues of special relevance for the tuning of data mining models, namely resampling strategies and oversearching.