Landmark Multidimensional Scaling in KNIME
Tabitha Goodall · 2012
This paper covers the introduction of Landmark Multidimensional Scaling (LMDS) to the KNIME data mining tool. It introduces KNIME, the MDS and LMDS algorithms, and projection quality assessment methods. Two KNIME nodes are developed: one implements LMDS, and one analyses the projection using internal validity measures to determine how well the original data structure is preserved. LMDS is tested using a visual assessment of the projection, numerical comparison of results, and execution time analysis. It is found to be of comparable accuracy visually and numerically to PCA when a good landmark set is chosen and its execution time when processing large data sets is significantly faster than that of PCA in all cases.