Clustering techniques for datasets with inter-cluster density variations

In unsupervised learning method, a big issue is formation of clusters even if the data does not contain clusters. If an clustering algorithm applies to a dataset it blindly divide the data into clusters because the algorithm is intended to do that. Hence before selecting any clustering method, it is better to analyze data whether it contains meaningful clusters or not. LS-VAT and LS-iVAT algorithms specify the number of clusters found in a dataset and show if there are existing clusters within clusters by creating a densely black square along the left diagonal of a square-size map. LS-VAT and LS-iVAT algorithms do not fit large datasets. To overcome the problems faced by LS-VAT and LS-iVAT algorithms, we propose Spectral Locally Scaled VAT (Spectral-LS-VAT) and Spectral Locally Scaled iVAT(Spectral-LS-iVAT) algorithms. Spectral-LS-VAT and Spectral-LS-iVAT algorithms produce better images for large datasets.