Spectral clustering based on extended deep ensemble auto encoder with eagle strategy

As an exploratory data analysis (EDA) process, spectral clustering (SC) reduces complex, multidimensional data sets to similar ones in rarer dimensions. Given the big challenges of high computational complexity and lack of accurate mapping in multidimensional data sets, it is essential to provide innovative solutions for SC. Against this background, the present study aims to propose a novel method based on extended Deep learning (DL), recruiting autoencoder (AE) ensembles. Here, the eagle strategy (ES) is further applied, rather than the anchor-based and landmark ones, to eliminate some complexities in SC and fill gaps in data training. As SC has a stochastic optimization structure, the ES may seem appropriate. The Fundamental Clustering and Projection Suite (FCPS) data set correspondingly represents that the proposed method has been able to surpass previous robust algorithms in terms of mean squared error (MSE), accuracy, adjusted random index (ARI), and normalized mutual information (NMI).

Automated summary

This study proposes a novel method based on extended deep learning and autoencoder ensembles, using the eagle strategy to address complexities in spectral clustering and data training gaps. Experimental results demonstrate that the proposed method outperforms previous algorithms on multiple metrics.