An Adversarial Regularized Deep Learning Framework for Clustering High Dimensional Data in the Big Data Cloud Paradigm

Abstract

Deep learning is suggested for the anonymity-preserving analysis of massive amounts of data. This technique converts the sensitive portion of personal data into non-sensitive data. A two-stage design is advised to finish this process. The approach makes use of CNN models and a modified sparse denoising autoencoder. CNN transforms the data and then classifies it using a modified sparse denoising autoencoder. Adding the sparsification parameter to the autoencoder's objective function using the Kullback–Leibler divergence function results in low loss in data transformation. In this case, the MSE (mean squared error) loss function is used to evaluate the model's efficacy. Three classes—Black (0), White (1), and Grey (2)—are created from the restored data. The deep CNN approach uses the characteristics from the sparse denoising autoencoder technique as input. This enables evaluation of the transformation process' correctness. Since the Black class data was transformed into Grey class data here, the CNN algorithm correctly identified the Black class data as Grey class during the classification stage with an accuracy of 0.99. The suggested approach works better than the current conventional methodologies, according to tests conducted using Cleveland medical datasets from the Skoda, Heart Disease, and Arrhythmia databases. A comparison between a simple autoencoder and the suggested method is given.