Design an Artificial Intelligence based Self-KYC Customer Onboarding System

https://doi.org/10.34047/mmr.v13.i2.p7

Abstract

Self-KYC allows telecom consumers to undergo identity verification remotely. Yet a journal-grade solution must include replicable methodology, explicit datasets, verifiable performance and a justified comparison with established baselines. This study proposes an end-to-end artificial-intelligence framework including identity-document processing, facial pre-processing, VGG16-based deep feature extraction, Multi-Scale Adaptive African Vulture Optimisation Algorithm (MS-AVOA) and Fully Connected Neural Network (FCNN) classifier. The evaluation is based on training procedures generated from LFW, CFP-FP, AgeDB-30 and large scale MS1MV2/MS1MV3 as presented in the thesis. The suggested model obtains 99.84% on LFW, 97.89% on CFP-FP and 97.41% on AgeDB-30. MS-AVOA improves the reported LFW accuracy from 97.40% for the original AVOA setup to 99.84%. Ablation studies demonstrate a noticeable decrease when optimization, face alignment, normalisation or augmentation are eliminated. The full pipeline reports an end-to-end latency of 0.279s, enabling near real-time onboarding. Statistical comparisons with AVOA, WOA, GOA and CSO give p-values smaller than 0.001. The results show that the combination of robust deep embeddings with adaptive feature optimization can increase the identity-verification accuracy, stability and operational adaptability to the telecom Self-KYC. Limitations around dataset provenance, liveness integration, demographic auditing, and validation at production scale are also discussed, for which solutions still need to be found before regulatory deployment.