Analisis Cluster Data Interkomparasi Anak Timbangan dengan Algoritma Self Organizing Maps

Abstrak

Intercomparison was conducted to determine the ability and the performance of the laboratory. Intercomparison results are usually expressed in the range of En ratio values (En ?|1|) which express the equivalence of one laboratory with other laboratories. If the laboratory is declared unequal, then it needs to identify the source of the problem by itself. To make it easier, it can be done by Clustering which is one of the data mining techniques. Clustering is done by applying a self organizing map algorithm on the KNIME (Konstanz Information Miner) analytic tools. Several experiments were carried out with different layer size and data normalization status from one experiment to another experiment. The results were analyzed through pseudo F statistical test and icdrate test. The largest pseudo F statistic value was obtained from the 8th experiment (setting the layer size 2x2 without data normalization) with a pseudo F statistic value of 167.53 for 1kg artifacts and a Pseudo F statistic value of 104.86 for 200 g artifacts where the optimum number of clusters are 4. The smallest icdrate value was obtained from the 5th experiment (setting the 2x3 layer size without data normalization) with an icdrate value of 0.0713 for 1kg artifacts and icdrate value of 0.2889 for 200g artifacts with the best number of clusters being 6. From 12 laboratories can be grouped into 6 groups where each group has the same identification. There are groups 1, 3 and 6 have 1 member, while groups 2, 4 and 5 have 3 members.