Contents I. Introduction
Data storage, analysis, and sharing are the essential services required by any organization to upgrade its performance [1]. Most of the businesses have shifted to the cloud due to its several benefits such as minimum upfront cost and maximum scalability for the required services [2]. However, once the data is transferred for storage and computation purposes in the cloud, the owners lose control over their data [3]. Multiple entities may access the data for commercial and/or other purposes after the data is outsourced [4]. It is not possible to fully trust the cloud platform because it is handled by the third party [5]. Therefore, before uploading data onto the cloud, owners first encrypt their data for privacy reasons. Although some conventional encryption techniques are available for the encryption of owners’ data, such as symmetric and fully homomorphic cryptography, these techniques are insufficient [6], [7]. However, it becomes difficult to perform the computation over the encrypted data [8]. There arises a necessity to protect the owners’ as well as the cloud data while performing the computation effectively. Furthermore, the stored and analyzed data must be shared with the various stakeholders to improve its utility. Although the data is shared among authorized entities, it cannot be assured that data will not be leaked by the receiving entities after obtaining it [9]. Thus, it is essential to protect the data from the entities involved in the communication process. To solve the above-mentioned challenges, we need an effective access control method that supports both the privacy and security of the owners’ data. To the best of the author's knowledge, no model exists that solves all the aforementioned challenges. In this regard, we propose a novel Machine Learning and Probabilistic Analysis based Model (MLPAM) for data protection through privacy-preserving data storage and analysis, secure sharing, and identification of guilty entity against data leakage in the cloud environment. The main contributions of MLPAM are summarized as follows.
To protect the data with enhanced security, all the entities are considered to be untrusted and MLPAM deals with involved entities by effectively defining an access policy.
MLPAM enables multiple data owners to freely share the outsourced data. In order to protect the data from stealing or leakage, the data of each owner is encrypted with a separate key and shared in encrypted form.
MLPAM uses two clouds where cloud1 deals with data storage, handling, and sharing whereas cloud2 generates the key for the encryption of owners’ data and performs the computation over the data obtained from cloud1 for privacy-preserving classification.
An effective distribution mechanism based on an access control is proposed for data distribution among multiple users, that enables to identify the guilty entity and reduces the risk associated with further leakage.
A series of experiments are conducted using the widely adopted datasets by researchers to validate the practicality of the proposed model. In addition to this, the comparisons are interpreted among the various a) datasets, b) classifiers, and c) distinctly preprocessed data using -differential privacy and with the state of the artworks to prove the superiority of MLPAM.
