Background: Machine learning (ML) methods are prevalent forecasting model construction tools that outperform conventional methods. This study is a systematic review of machine learning method utilization for load and energy consumption forecasting between 2020-2025. The study covers a variety of methods, ranging from simple algorithms such as linear regression and support vector machines to complex deep learning models such as LSTM, Convolutional Neural Networks CNNs, Transformer models, Graph Neural Networks GNNs, and particular ensemble and hybrid methods. Methods: This study systematically reviewed electric load and energy demand forecasting machine learning techniques with strict methods and harvested primary research databases and preprint servers for English-language papers from January 2020 to May 2025. Results: This study revealed that deep learning models, including LSTM and CNN-LSTM, are becoming more widely used, which indicates a shift towards operational maturity. However, their complexity can be difficult for low-resource environments. The performance of Machine learning models is vastly context dependent. It is a function of factors such as the size, resolution, and range of forecasting involved, thus requiring the proper selection of models. Above all, quality data and proper pre-processing always prevail over the effect of selected machine learning techniques. Conclusion: Machine learning has assisted energy forecasting a lot but falls short on usability and reliability. More technology and collaboration are required to succeed with renewable energy systems. Originality/Novelty of this article: This study describes new developments in Machine learning for energy forecasting and mentions trends and issues to be expected. It recommends what is in the pipeline for future research and applications.
Akinrogunde et al. (Thu,) studied this question.