A NOVEL PRIVACY PRESERVING BASED ENSEMBLE DEEP LEARNING FRAMEWORK FOR LARGE DIMENSIONAL MEDICAL DATASETS

Abstract

With the rapid growth of data size, computing power and high dimensionality, it is essential to implement a novel privacy preserving model in deep learning framework. However, most of the traditional machine learning applications are integrated in deep learning framework on large databases, it is essential to secure the large sensitive patterns that are generated by the machine learning approaches before uploading to the cloud storage. As a result, how to design and implement a novel privacy preserving deep learning model (PPDLM) over high dimensional cloud data becomes a challenging task. Traditional privacy preserving deep learning frameworks are depends on data transformation approaches rather than the cryptographic approach due to high computational memory and time in cloud computing. In the real time multi-user applications, multiple datasets are distributed across the multiple users for privacy preserving. As the data size of multi-user applications increases, traditional PPDLM models require high computation memory and time for preserving the machine learning patterns. To overcome these problems, a novel data partitioning based privacy Preservation deep learning model is implemented on high dimensional datasets. Experimental results proved that the present system has high computational accuracy with privacy in the patterns compared to the existing models.