Conventional spectral clustering methods provide essential information about the structure within the dataset; nonetheless, they are not scalable over large datasets. In this research, an ensemble of density-based and cluster-based sampling techniques is used to improve the scalability of spectral clustering in a novel way. By carefully choosing a representative sample of data points, the suggested strategy lowers computing complexity while speeding up the clustering process without sacrificing accuracy. Our tests show that the suggested approach performs better in terms of clustering performance (as determined by the Silhouette Score, Adjusted Rand Index, and Normalized Mutual Information) and computing efficiency than conventional spectral clustering and other cutting-edge approaches. In large-scale datasets, the approach reduces execution time by 134.4% while improving clustering accuracy by 61.5%. This method is potentially used in large-scale data analysis applications where scalability and efficiency are crucial, including body area networks (BANs).