Comprehensive Evaluation of Regression Algorithms for Power Grid Datasets: Effectiveness and Robustness Analysis

As power grid systems increasingly undergo digital transformation, their data management and analysis become substantially more complex, characterized by large-scale datasets, intricate spatio-temporal relationships, and frequent disturbances such as noise, data missingness, and anomalies. Accurate and robust regression models thus play an essential role in enabling reliable real-time monitoring, predictive maintenance, load forecasting, and operational decision-making within power systems. Despite extensive research into individual regression techniques applied across various fields, comprehensive comparative evaluations focusing explicitly on the performance and robustness of diverse regression methods for power grid datasets remain scarce. This study systematically evaluates and contrasts extensive representative regression algorithms across real-world benchmarks and typical power grid scenarios, where the data size varies from hundreds to hundreds of thousands of samples and feature sizes range from ten to thousands. We not only assess the effectiveness of regression models in terms of prediction precision but also investigate their robustness against various data disturbances with different levels of Gaussian noise and missing values. These findings offer valuable insights into the strengths and limitations of each regression technique, thereby providing clear and actionable guidelines for power system engineers and data scientists in selecting optimal regression approaches tailored to specific operational scenarios and reliability requirements.