[Paper Review] Privacy-Preserving and Outsourced Multi-User k-Means Clustering

Bharath K. Samanthula, Fang-Yu Rao+3 more|arXiv (Cornell University)|Dec 14, 2014

Privacy-Preserving Technologies in Data2 references4 citations

TL;DR

This paper proposes a novel privacy-preserving and outsourced multi-user k-means clustering protocol (PPODC) that enables n users to securely outsource their encrypted data to a federated cloud environment for collaborative clustering. By eliminating secure division operations through an efficient transformation technique, the protocol ensures data confidentiality under the semi-honest model while incurring negligible computation cost on users, with cloud-side computation linear in data size, attributes, and clusters.

ABSTRACT

Many techniques for privacy-preserving data mining (PPDM) have been investigated over the past decade. Often, the entities involved in the data mining process are end-users or organizations with limited computing and storage resources. As a result, such entities may want to refrain from participating in the PPDM process. To overcome this issue and to take many other benefits of cloud computing, outsourcing PPDM tasks to the cloud environment has recently gained special attention. We consider the scenario where n entities outsource their databases (in encrypted format) to the cloud and ask the cloud to perform the clustering task on their combined data in a privacy-preserving manner. We term such a process as privacy-preserving and outsourced distributed clustering (PPODC). In this paper, we propose a novel and efficient solution to the PPODC problem based on k-means clustering algorithm. The main novelty of our solution lies in avoiding the secure division operations required in computing cluster centers altogether through an efficient transformation technique. Our solution builds the clusters securely in an iterative fashion and returns the final cluster centers to all entities when a pre-determined termination condition holds. The proposed solution protects data confidentiality of all the participating entities under the standard semi-honest model. To the best of our knowledge, ours is the first work to discuss and propose a comprehensive solution to the PPODC problem that incurs negligible cost on the participating entities. We theoretically estimate both the computation and communication costs of the proposed protocol and also demonstrate its practical value through experiments on a real dataset.

Motivation & Objective

• Address the challenge of privacy-preserving distributed clustering (PPDC) in resource-constrained environments where users lack sufficient computational power.
• Overcome the high communication and computation costs of existing PPDC solutions by enabling secure outsourcing of clustering tasks to the cloud.
• Develop a practical, efficient, and privacy-preserving solution for multi-user k-means clustering under a federated cloud model with two untrusted cloud providers.
• Ensure data confidentiality of all users throughout the clustering process without requiring decryption at any point.
• Minimize user-side computational overhead to near-zero while maintaining strong privacy guarantees and scalability for large datasets.

Proposed method

• Utilize a federated cloud architecture with two untrusted cloud service providers (e.g., Amazon and Google) to jointly perform clustering on encrypted data.
• Apply a novel transformation technique to construct an order-preserving Euclidean distance function over encrypted data, avoiding the need for secure division operations.
• Design an iterative clustering protocol where cluster centers are computed securely using homomorphic encryption and secure comparison protocols.
• Implement a secure termination detection mechanism based on convergence of cluster centers, evaluated over encrypted data using threshold-based checks.
• Decompose the clustering process into stages: (1) secure data aggregation and distance computation, (2) cluster assignment, (3) secure center re-computation, and (4) convergence check.
• Optimize performance by enabling parallelization of record-level assignments in Stage 2, leveraging cloud-native parallel processing frameworks like Spark or Hadoop.

Experimental results

Research questions

• RQ1 How can k-means clustering be securely performed on encrypted data from multiple users without revealing any sensitive information?
• RQ2 What techniques can eliminate the need for secure division operations in privacy-preserving clustering, thereby reducing computational overhead?
• RQ3 Can a federated cloud model with two untrusted clouds enable efficient and private clustering while minimizing user-side computation?
• RQ4 What is the scalability and performance trade-off of the proposed protocol for large-scale datasets with varying numbers of records, attributes, and clusters?
• RQ5 To what extent can the protocol be optimized through parallelization of independent operations in the clustering pipeline?

Key findings

• The proposed PPODC protocol achieves data confidentiality for all users under the semi-honest threat model, ensuring no user data is exposed during computation.
• The protocol eliminates secure division operations via a novel transformation, significantly reducing computational complexity compared to prior approaches.
• Online computation time grows linearly with the number of data records (m), attributes (l), and clusters (k), with 36.14 minutes required for m=6,000, l=10, k=8.
• Over 99% of the total computation time is spent in Stage 2 (cluster assignment), which is highly parallelizable at the record level.
• User-side computation cost is negligible—measured in milliseconds—making the protocol practical for resource-constrained end-users.
• The protocol’s performance can be further improved via parallelization using cloud-native frameworks like Spark or Hadoop, enhancing scalability for big data workloads.