Preserving Data Integrity and Detecting Toxic Recordings in Machine Learning using Blockchain

Abstract

Machine Learning (ML) is receiving unprecedented hype and attention. However, the ML runtime environment is still at risk from threats, such as manipulation of model parameters or contradictory poisoning of training datasets. A blockchain is a technology that combines a set of existing techniques, protocols, and tools to form a distributed and secure ledger of all transactions. This article examines and proposes a way of integrating ML suitable for Blockchain to protect the training dataset and model parameters. Another major contribution of this work is the deployment and securing of the decision process of ML, Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP) models. This smart contract-based deployment has equipped the Blockchain-based system to detect toxic recordings intelligently. The effectiveness of this proposed approach is measured in both its detection capabilities and its operational efficiency, by applying a case study of medical records as a sensitive area that tested the performance of this approach.