Automated recognition of construction activities is central to real-time occupational safety and health (OSH) monitoring. However, integrating machine learning (ML) with blockchain to strengthen privacy and cybersecurity may reduce computational efficiency. To balance this trade-off, this study proposes an efficiency-enhanced machine learning on blockchain (MLOB) framework designed to support real-time performance while maintaining security requirements.
The efficiency-enhanced MLOB framework combines three optimizations: (1) model distillation to reduce ML complexity, (2) parallelization to distribute inference and verification workloads and (3) blockchain configuration tuning to improve consensus and transaction handling. Performance is assessed on a real-world construction activity recognition task and benchmarked against a baseline implementation using end-to-end latency, throughput and security metrics.
Relative to the baseline, MLOB reduces end-to-end latency by 48.4% and increases throughput by 205.8%, while maintaining security performance. These gains enable near real-time, privacy-preserving decision support for OSH applications.
The study presents a novel, integrated MLOB architecture that jointly optimizes the ML and blockchain parts. It offers an implementation-ready blueprint for scalable, secure and time-critical construction safety analytics.
