Читать книгу Data Analytics in Bioinformatics - Группа авторов - Страница 41

Generally gene expression data is nonlinearly separable and the gene data is represented in a matrix format using linear kernel function as shown in Figure 2.2. IKKM is applied on the kernel matrix in which clusters need not be predefined.

The algorithmic approach of IKKM is carried by calculating the centre of mass of the entire dataset and then identifies the element (×1) which is distant from the centre of mass (C) another element (×2) is calculated distant from c1 is calculated. Now the elements near to c1 are moved into cluster (S1) and the elements near to c2 are moved into another cluster (S2). This process is repeated until no change is observed in the clusters. In contrast to traditional k-means approach this algorithm implementation doesn’t require predicting the number of clusters in advance.

Drawback—Cannot deal with high dimensional dataset [16].