Spectral clustering is one of the most relevant unsupervised learning methods capable of classifying data without any a priori information. At the heart of this method is the computation of the dominant eigenvectors of an affinity matrix in order to work on a low-dimensional data space in which the clustering is made. We propose in this paper a study of the integration of the FEAST library to compute these eigenvectors in our parallel spectral clustering method by domain decomposition. We also show that this library allows to add a second level of parallelism in addition to the domain decomposition level.