Optimized smart grid fault detection model using gradient boosting machines

The evolution of traditional power grids into intelligent, resilient infrastructures has become imperative to address growing energy demands, climate-induced disruptions, and the integration of renewable energy sources. This study presents an AI-enhanced smart grid framework that employs machine learning models to optimize energy forecasting and fault detection, thereby improving grid reliability and operational efficiency. Specifically, the study implements a Gradient Boosting Regressor (GBR) for short-term load forecasting and a Gradient Boosting Classifier (GBC) for real-time fault detection. A balanced dataset, derived through oversampling techniques, ensures robust model training and classification reliability. Experimental results from simulated grid data demonstrate high performance, with the forecasting model achieving a coefficient of determination (R²) of 0.93 and low prediction errors (RMSE = 12.08, MAE = 9.37). The fault detection model attained 96.1% accuracy, 93% precision, and 100% recall for fault classification, resulting in an F1-score of 0.96, comparable or superior to benchmarks in the literature. These results validate the proposed system’s suitability for implementation in developing regions, particularly in Sub-Saharan Africa, where grid instability and outage frequency hinder socioeconomic development. By integrating real-time predictions with edge-level intelligence, this research contributes a scalable and context-aware solution to modernize energy systems in underserved environments. The study concludes by recommending policy and technological pathways for localized adoption of AI in power distribution networks.