k-eNSC: k-estimation for Normalized Spectral Clustering

In machine learning, the power of an approach is measured by its capability to be adapted for different applications and using different formats of data. Spectral Clustering is an unsupervised method that can be adopted for many research fields in and beyond computer science. In this paper, we present and analyze the existing algorithms of spectral clustering, and based on their limits we propose our modified version to deal with the most common challenge in this context which is the dynamic estimation of the output number of clusters.