A geospatial machine-learning framework for regional electrification estimation mapping and trend analysis using demographic and infrastructure features

Abstract

Many rural communities in Ethiopia's Amhara region lack electricity access despite substantial urban-rural disparities. Traditional surveys are costly and inconsistent, while Nighttime Light (NTL)-population threshold methods suffer from arbitrary thresholds, off-grid detection failures, and light-access conflation. This study develops a geospatial machine learning framework using Visible Infrared Imaging Radiometer Suite-Day/Night Band (VIIRS-DNB) nighttime lights, population density, and distances to grid lines and roads. Random Forest (RF) outperforms Support Vector Machine (SVM) and Decision Tree (DT), achieving 86.8% overall accuracy and a 71.3% Kappa coefficient for operational mapping of electrification status and estimation of regional electrification rates from 2018 to 2024. Validation against GPS-recorded transformer locations demonstrates that the Random Forest model reconstructs a realistic, corridor-type electrified network and outperforms a calibrated NTL-population threshold approach, which tends to miss grid-connected rural corridors and overestimate urban brightness effects. The resulting maps reveal zone-level disparities and steady growth to 41% electrification by 2024, with population density (27.49% importance) emerging as the leading predictor among balanced variables. This framework supports SDG 7 planning and investment prioritization in underserved areas.