High dimensional data clustering have gained significant attention recently due to increased utilization of the high dimensional data streams across the large distributed cloud. High dimensional data clustering approaches has been developed in the conventional work using both machine learning and deep learning architectures. Despite of many advantageous, it is inefficient in handling the curse of dimensionality and data sparsity issues. Further learning model leads to high computational complexity and data disclosure attacks to various transaction queries to cloud servers. In order to mitigate those challenges, high scalable secure regularized model has been designed which is entitled as Deep Adversarial Regularized Autoencoder Technique. Autoencoder is a popular mechanism to accomplish dimensionality reduction. Proposed model deeply explore the latent structure of the data and computes the associations of the data points to construct the spatial and temporal cluster structures to high dimensional data as new clustering perspective. It discriminate the data points efficiently. Further evolving data streams are approximated using the variational autoencoder to preserve the cluster structures. Euclidean distance is employed in embedding function of the autoencoder to generate the efficient data clusters with minimized intra cluster similarity and inter cluster variation in the feature space. Hyper parameter tuning using RMSProp has been enabled in the output layer to make the data instance in the cluster to be close to each other by determining the affinity of the data on new representation. Experimental analysis has been performed on benchmarks datasets such as Twitter and Forest Cover to compute the proposed model performance with conventional approaches. The performance outcome represents that good scalability and effectiveness on high dimensional data has been reached.

High Dimensional Data clustering, Big Data Cloud, Privacy Preserving, Variational Autoencoder, Advanced Networking