Hierarchical Clustering Algorithm Based on Density Peaks using Kernel Function for Thalassemia Classification

Abstract

Thalassemia is an inherited blood disorder and one of the most common genetic diseases in the world, with its classification due to the processes involved in patients' treatment. Therefore, it is essential to obtain a precise diagnosis due to an appropriate treatment receives a higher lifetime value. A new method based on the kernel, modified from hierarchical clustering based on density peaks (HCDP) was proposed in this research. Furthermore, the concept of k-nearest neighbor and hierarchical clustering, consists of three steps were utilized. These steps include local density calculation, hierarchy representation, and optimal cluster extraction. The modification of this technique is therefore based on the polynomial kernel function, which was expected to be more accurate in separating data that cannot be detached linearly. Data utilized was obtained from Harapan Kita Hospital, West Jakarta, Indonesia, and it consists of 82 thalassemia and 68 non-thalassemia samples. Using the proposed method, the performance of HCDP with/without kernel function in this paper were examined using 10-fold cross-validation and compared using the confusion matrix by calculating its F1-Score. The results concluded that hierarchical clustering based on density peaks gives approximately 67.77 percent F1-Score, while 70.06 percent is obtained when the method is combined with the kernel function.