Tuned Data Mining (TDM) and TDMR
The complex, often redundant and noisy data of real-world data mining (DM) applications frequently lead to inferior results when out-of-the-box DM models are applied. Tuning of parameters is essential to achieve high-quality results. We pursue in this project an approach to tune parameters of the preprocessing and the modelling phase conjointly. We propose the new framework TDM (Tuned Data Mining) which facilitates the search for good parameters and the comparison of different tuners by using mostly generic elements which are easily applied to new tasks.
The R-package TDMR (Tuned Data Mining in R) - freely available as open-source software from CRAN - is written with the aim to facilitate the setup, training and evaluation of data mining models. It puts special emphasis on tuning these data mining models as well as simultaneously tuning certain preprocessing options. TDMR is especially designed to work with SPOT as the preferred tuner, but it offers also the possibility to use other tuners (CMA-ES, LHD, direct-search optimizers) for comparison.
See the user manual TDMR-docu.pdf (CIOP-Report [Kone12a], last update March 2015) and the TDMR Tutorial (CIOP-Report 03/2012 [Kone12b], last update March 2015) for in-depth information on usage and development of the TDMR package.
|Dr. Patrick Koch, FH Köln||Prof. Dr. Wolfgang Konen, FH Köln|
Publications in the area of Tuned Data Mining (TDM) and TDMR: