Integrated Feasibility Assessment of Edge, Cloud, and Hybrid AI-as-a-Service Architectures for Face Recognition

Deploying face recognition as a service requires evidence beyond recognition accuracy because architecture affects latency, scalability, cost, and biometric-data exposure. This study evaluates Edge, Cloud, and Hybrid architectures through an integrated technical, techno-economic, and regulatory-readiness framework. EdgeCloudSim experiments modeled 100–500 connected devices using an infrastructure-level workload abstraction structurally mapped to published face-recognition pipelines, while explicitly distinguishing this abstraction from direct neural-inference benchmarking. The technical screening thresholds were treated as scenario-specific engineering gates and evaluated through threshold sensitivity analysis. A five-year comparative cash-flow model applied the same modeled avoided operating-cost benefit to all architectures. At 500 devices, Hybrid produced 2.2066 s processing time, 16.736% failed tasks, and 0.1355 s total latency; its failure rate was lower than Edge (27.797%) and Cloud (22.137%). Under the common economic-benefit scenario, Edge and Cloud generated negative NPVs of IDR 982.50 million and IDR 247.25 million, respectively, whereas Hybrid generated an NPV of IDR 958.96 million, IRR of 30.73%, PI of 1.64, and an actual cumulative payback period of 2.60 years. Regulatory mapping assessed implementation readiness rather than demonstrated legal compliance. Hybrid therefore remains the most balanced conditional baseline under the stated assumptions, subject to empirical pipeline calibration, capacity upgrades, privacy controls, and field validation.