The comparison of k-modes clustering and ROCK clustering to the poverty indicator in Samadua Subdistrict, South Aceh

The cluster methods including k-modes clustering and ROCK clustering are rarely used for analyzing categorical data. Much of the research data, however, are categorical, such as the poverty indicator dataset. Many studies have examined each of these two methods, but there is still limited study for comparing them. This study compared between k-modes clustering and ROCK clustering on a dataset of poverty indicators at a category scale. There were 1449 households with the lowest 40% income in Samadua Sub-district, South Aceh Regency were used as data in this study. The result of this research was that the optimal number of clusters for k-modes clustering was 2. Similarly, the optimal number of clusters for ROCK clustering was 2 with a threshold value (θ) of 0.29. The SW/SB and R-Squared values for the k-modes clustering were 0.0149 and 0.7558, respectively, compared to 0.0682 and 0.5732 for ROCK clustering. From these two values, it can be concluded that the k-modes clustering is better than ROCK clustering. The significant difference between the two clusters based on deciles of households with the lowest 40% income of the k-modes clustering is ownership of defecation facilities, ownership of television and motorbikes.