Data-Centric AI Engineering For Reliable KPI Prediction And Anomaly Diagnosis In Telecommunications

Abstract

This article presents a data-centric AI engineering framework for telecommunications networks that positions data quality as the primary driver of performance in Key Performance Indicator (KPI) prediction and anomaly detection. The proposed framework "5D Model"-Define, Diagnose, Design, Derive, Deploy—addresses persistent challenges in telecom datasets, including noise, sparsity, inconsistency, and multi-vendor heterogeneity. Empirical benchmarking across 8 major carriers demonstrates that data-centric interventions yield 2.6-4.0x greater performance improvements than architecture enhancements across diverse network environments.