Year: 2025
Type: article
Abstract: Serverless computing is increasingly integrated with artificial intelligence to address scalability, cost inefficiency, and operational complexity in conventional cloud-based machine learning workflows. This investigation evaluates the performance, scalability, cost efficiency, and statistical robustness of serverless AI architectures for cloud-based machine learning workflows under controlled academic settings. An experimental study was conducted at the Department of Computer Science and Engineering, Daffodil International University, Dhaka, Bangladesh, from June to December 2024. Thirty cloud-based machine learning workflows were implemented using serverless architectures. Performance metrics included latency, execution time, throughput, scalability index, resource utilization, and cost efficiency. Statistical analysis employed mean values, standard deviation, and hypothesis testing with significance set at p<0.05. Serverless AI workflows demonstrated a 42.6% reduction in mean execution time compared with baseline container-based systems (3.8±0.9 s vs. 6.6±1.2 s; p=0.001). Average operational cost decreased by 38.4%, with per-inference cost reduced from USD 0.021±0.006 to USD 0.013±0.004 (p=0.003). Scalability efficiency improved by 51.2%, enabling automatic handling of workload surges up to 280% without performance degradation. Mean latency decreased by 34.7% (p=0.008). Resource utilization variance was significantly lower (SD 0.41 vs. 0.88), indicating improved stability. Overall system reliability increased to 96.3%, compared with 88.5% in non-serverless deployments. Serverless AI significantly enhances efficiency, scalability, and cost-effectiveness of cloud-based machine learning workflows, supporting its adoption as a resilient and statistically robust computational paradigm.
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7124729020
