Comparative Analysis of Machine Learning Models for Short-Term Electricity load Forecasting: A Comparative Review of the Literature

Abstract

This paper presented a comparative literature review of machine learning models for short-term electricity load forecasting, with emphasis on Artificial Neural Network (ANN), Random Forest (RF), Support Vector Machine (SVM), Long Short-Term Memory (LSTM), and hybrid CNN-LSTM models. Motivated by the increasing operational challenges associated with inaccurate electricity demand prediction, particularly within developing power systems characterized by nonlinear consumption patterns, demand variability, and infrastructural instability, this paper adopted a comparative analytical approach based on recent empirical studies focusing on forecasting accuracy, computational efficiency, robustness, scalability, and adaptability of machine learning techniques in electricity load prediction. The results, tables, and figures presented in this paper are drawn from and evaluated findings reported in the cited secondary literature rather than from an original experiment conducted by the authors. Findings from the reviewed studies, tables, and visual analyses revealed significant performance variations among the forecasting models. Comparative findings presented in the performance evaluation tables and error analysis charts showed that LSTM and hybrid CNN-LSTM architectures consistently achieved the lowest forecasting error values among the models reviewed, with reported RMSE and MAPE reductions relative to the other benchmark models. The comparative bar charts and performance ranking figures further indicated that LSTM and CNN-LSTM demonstrated superior capability in learning nonlinear, sequential temporal load patterns, while Random Forest exhibited stronger robustness and forecast stability under noisy and incomplete datasets, and ANN showed competitive but comparatively weaker performance than the deep-learning architectures. Heatmap visualizations additionally revealed strong positive correlations between historical load variables, temporal indicators, and forecast accuracy. The paper concluded that machine learning techniques provide reliable and adaptive solutions for short-term electricity demand forecasting. It is recommended that electricity distribution companies adopt LSTM and hybrid CNN-LSTM-driven predictive systems complemented by Random Forest for operational contexts requiring greater robustness under noisy data, integrated with real-time monitoring infrastructures. Future research should explore hybrid deep learning architectures, smart-grid forecasting integration, renewable energy forecasting synchronization, and explainable artificial intelligence techniques for interpretable energy forecasting systems.