Short-term load forecasting (STLF) is central to ensuring reliable electricity supply and economic operation of power systems. However, its accuracy depends heavily on the forecasting methodology employed, particularly in area of much load consumptions and commercial activities like Aba. This study presents a comparative evaluation of four distinct forecasting approaches linear regression (LR), feedforward neural networks (NN), Elman recurrent neural networks (ERNN), and a hybrid ensemble learner applied to historical data Geometrics Integrated Energy Services, located in Aba, Abia State, Nigeria. Using hourly load, amperage, and weather data collected between September 2024 and March 2025, the models were tested on both historical (Weeks 38–39) and prospective future (Week 14) horizons. Performance was assessed using root-mean-squared error (RMSE), mean absolute error (MAE), Mean Absolute Percentage Error (MAPE), Symmetric Mean Absolute Percentage Error (SMAPE), and R² metrics. Results show that the hybrid model achieved the highest accuracy (RMSE = 0.2608, R² = 0.7079), followed by the neural network (RMSE = 0.3720, R² = 0.4056), and linear regression (RMSE = 0.4200, R² = 0.2420). The ERNN model performed poorly (RMSE = 0.5683, R² = –0.3879), reflecting its limited adaptability to constantly changing load patterns in areas feeding from the Geometric. The comparative analysis not only establishes the relative strengths and weaknesses of each model but also provides practical guidance for utilities in selecting forecasting techniques appropriate for local grid conditions. The study concludes that while traditional and recurrent approaches remain valuable, ensemble learning offers the most balanced and reliable forecasts for emerging electricity markets.