Cluster Evaluation Weighing Intercomparison Data with Self Organizing Maps Algorithm

Abstract

Laboratory intercomparison is one of method to determine the ability and assess the performance of a laboratory. Laboratory performance can be seen from the evaluation of the En ratio’s value, which is a comparison between the difference in the value test of the participant's laboratory with reference’s laboratory and the difference in the square root of the uncertainty value form participant's laboratory and reference’s laboratory. The laboratory is declared equivalent if the En value is in the range of En ≤|1|. Intercomparisons evaluation can also be done by utilizing one of the data mining technologies in the form of clustering. Clustering is done by using self-organizing maps algorithm, which is an unsupervised learning algorithm. The advantage of clustering in evaluating intercomparation data lies in its ability to group data into several clusters that have closeness or similarity in characteristics / traits / characters of data, making it easier for intercomparation organizers to provide analytical recommendations for improving laboratory performance. Intercomparation data are grouped based on the homogeneity between members in one cluster and heterogeneity among the clusters. To get the best number of clusters, evaluation is carried out through three testing methods, pseudo-F statistic, icdrate and davies bouldin index. From several experiments, the largest pseudo-F statistic value was 167.53, the smallest icdrate value was 0.071 and the smallest DBI value was 0.053 for the 1000 g artifact. As for the 200 g artifact, the largest pseudo-F statistic value was 104.86, the smallest icdrate value was 0.289 and the smallest DBI value was 0.306