Studies of Self-Organizing Maps (SOM) In Grouping Objects (case study: grouping of villages/urbans in Wajo Regency, South Sulawesi).

Abstract

Clustering is a process of classifying objects into groups which have similarity. The result of clustering will show that objects in one cluster will be more homogeneous than others. There are two methods in classic clustering analysis i.e. hierarchical cluster method and non-hierarchical cluster method. Determination of the number of clusters which formed by them is done subjectively. The cluster other methods also developed by using artificial intelligence. Artificial neural network is an information processing paradigm that inspired by the biology systems, it is neuron. Like brain which process information. Self-organizing maps (SOM) is one of the topology of Unsupervised Artificial Neural Network (Unsupervised ANN) which process does not require monitoring in his training. Application clustering using SOM algorithm is expected to be used as a tool to analyze the data in order to obtain the characteristics of the data that will be grouped. Clustering is used to group the data naturally without based on the specific class target. In this study, SOM compared with clustering method with large data sizes, it was two-step cluster. According to Bacher (2004), two-step cluster method (TSC) was a cluster method which can resolve the problem clustering measurement scale, especially for large data with variables which have categorical and numerical data types. Performance clustering SOM and two-step cluster method compared by the simulation data, afterwards, applying the method of SOM on clustering villages/urbans in Wajo regency, South Sulawesi. The data in this study consisted of two sources i.e. simulated data and secondary data. Simulated data was generated data multivariate distribution (μ,Ʃ) which useful to measure the performance of two-step cluster method and SOM in classifying an object. Secondary data, which used in this study, BPS’s data in Wajo regency, South Sulawesi, was Village Potential Data (VPD) in 2011. Simulation data was the generated data numeric type (M) which consisted of three forms of the population i.e. a). a population consisted of three clusters were clearly separated, b). a population which consisted of three clusters of overlapping (overlap) each other in small numbers, and c). a population that consisted of three clusters of overlapping (overlap) each other in large numbers. The results of methods SOM and TSC showed that simulation data has the good ability to classify data, however, TSC provides better clustering results for large data sizes than SOM. In addition, it is also showed that the larger the number of data, the misclassification of SOM would become larger, nevertheless, the changes were relatively smaller. In the other hand, the larger the number of data the misclassification of TSC method was become smaller. Secondary data, which used in this study, written documentation and identification of used variables about areas/villages in Wajo regency, South Sulawesi, was Village Potential Data (VPD) in 2011 i.e. : X1 (total population), X2 (family farm), X3 (family farm laborer), X4 (family power user), X5 (fuel for daily cooking), X6 (educational facilities), X7 (health personnel), X8 (population mortality). Objects in this research were applied around the village/urbans in Wajo district. The results of clustering with SOM method, village/urbans in the Wajo regency produced 3 clusters. The formed clusters have the same descriptive value between clusters, and each clusters which formed, was dominated by a few districts in that area. Some other villages/urbanss incorporated also have similar caracteristic of variables, e.g. they being around the district.