Machine Learning for Forecasting Electricity Demand in Emerging Economies Amid COVID-19 Uncertainty

Predicting electricity demand in emerging economies is challenging as congested cities are exposed to unpredictable mobility patterns, weather sensitivity, and global pandemics such as COVID-19. Although mass public transportation systems have long supported urban mobility, limited understanding remains of how disruptions to these travel modes impact the energy sector. This paper introduces a framework that leverages metro ridership data, a point of interest, to enhance the predictive modeling of electricity demand. Convolutional neural networks (CNNs) are trained across various pandemic phases, enabling more effective mapping between multi-scale features and electricity demand. Results demonstrated that the model receiving historical data related to COVID-19 test positivity rates and metro users was more robust to the early impacts, achieving a MAPE of 3.35% and RMSE of 59.82. The research was validated through a case study of the Republic of Panama, featuring one of Latin America’s most modern rapid transit systems, which transports an average of 365,000 passengers daily to the city’s center as of 2025. A SHAP analysis further demonstrates that shifts in metro ridership patterns influenced model outcomes during the pandemic’s peak from 2020 to 2022.