Improved Fuzzy Possibilistic C-Means using Artificial Bee Colony for Clustering New Student’s Financial Capability to Determine Tuition Level

Outliers in the dataset will affect the quality of the cluster, so a good clustering method is needed. Based on the Mahalanobis distance method, it is known that the research dataset has outliers. Clustering methods that are often used for this type of data are Fuzzy C-means (FCM), Possibilistic C-means (PCM), and Fuzzy Possibilistic C-means (FPCM). This study aims to develop a clustering method that is more robust to outliers by using the Artificial Bee Colony (ABC) algorithm to minimize the objective function of FPCM. This study produces a new algorithm called Artificial Bee Colony Fuzzy Possibilistic C-Means (ABCFPCM) so that the resulting clusters are not easily trapped in the local optimum. This study also provides cluster centroid initialization using K-Means++ to improve cluster quality. ABCFPCM performs best because it significantly increases the Silhouette value and the Between Sum Squares (BSS) and Total Sum Squares (TSS) ratio. ABCFPCM performance provides the best cluster quality of 72.16% based on the BSS/TSS ratio, FPCM of 70.71%, and FCM K-Means++ of 68.14%. K-Means++ in the cluster method does not affect cluster performance except for FCM, where cluster quality is slightly increased. The centroid results of 8 clusters as the best performance of ABCFPCM are used to determine the tuition rate level. The impact of this study is to improve the quality of FPCM performance because it is no longer trapped in a local optimum at the cluster centroid.