Metode ROBPCA (Robust Principal Component Analysis) dan Clara (Clustering Large Area) pada Data dengan Outlier

Studi Kasus Data Laporan Indeks Kebahagiaan Dunia Tahun 2018

Abstrak

Principal Component Analysis (PCA) is one of multivariate analysis used for deputizing variables using less number of Principal Components without losing much information. In other words, it is used for explaining the underlying variance-covariance structure of the large data set of variables through a few linear combinations of these variables. PCA is significantly influenced by the outliers, since the covariant matrix are sensitive to outliers. Thus, the analysis for this study was conducted by using a PCA that is robust to outliers, namely ROBPCA or Hubert PCA. Then, the principal components formed were used as inputs in cluster analysis using the Clara method (Clustering Large Area). Clustering Large Area is one of the k-medoids methods that is robust to outliers and is appropriate for large data analysis. In the case study of the compiling variables of happiness index based on The World Happiness Report (WHR)2018 using the Clara method with Manhattan distance, the best average value of Overall Average Silhouette Width in the 5 clusters were obtained.