**5. Conclusions**

The goal of this study was to determine whether there is a statistically significant difference in the performance of various well-known simple machine learning (ML) models when they are applied to the prediction of power demand and supply. In order to accomplish this, six well-known machine learning methods were tested using data from the Eskom database, which included hourly system demand and renewable generation datasets. The ML algorithms considered include the artificial neural network, Gaussian regression, K-nearest neighbor, linear regression, random forest, and the support vector machine, among other methods of data analysis. Fairness was achieved by ensuring that the hyperparameters of each algorithm were fine tuned to the greatest degree possible. Our findings suggest that, within the confines of the datasets used in this study, there was little/no statistically significant difference between the different models in terms of both quantitative and qualitative measures, which is particularly noteworthy, given that they were all meticulously fine tuned. Additionally mentioned is the importance of reporting as many metrics as possible, particularly the correlation coefficient and absolute and squared errors, in order to ensure that fair conclusions are formed when comparing different machine learning algorithms. Based on the fact that each metric often reports a separate performance measure and that selective reporting may result in erroneous conclusions, this requirement is recommended. Furthermore, when it came to estimating the wind power generation dataset, all of the models performed poorly, which we attributed to the extremely stochastic nature of wind energy as a source of energy, as previously stated in the literature. This may imply that improved models for smart grid systems may be required, particularly in areas where wind power constitutes a significant portion of the generated electricity. In spite of this, it is possible that any ML model can still be used for power prediction in smart grid systems, particularly in situations where demand and generation follow regular patterns, and provided that the model's hyperparameters are properly tuned based on the type of input data being used. Finally, we stress that further robust investigations, particularly those based on the use of larger datasets from a wider range of sources, should be strongly encouraged in order to either substantiate or refute the conclusions of the present paper.

**Author Contributions:** These authors E.C., A.J.O. and S.J.I. contributed equally to this work. Conceptualization, A.J.O., E.C. and S.J.I.; methodology, E.C. and A.J.O.; writing—original draft preparation, A.J.O. and E.C.; writing—review and editing, A.J.O. and S.J.I.; supervision, A.J.O. and S.J.I.; funding acquisition, S.J.I. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the COUNCIL FOR SCIENTIFIC AND INDUSTRIAL RE-SEARCH (CSIR) with project number 05400 054AT KR2EEMG. and The APC was funded by project number 05400 054AT KR2EEMG.

**Institutional Review Board Statement:** Not applicable

**Informed Consent Statement:** Not applicable

**Data Availability Statement:** Not applicable

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.
