In March 2025, multiple large-scale wildfires simultaneously broke out across South Korea and rapidly spread to neighboring municipalities, resulting in the most extensive wildfire damage in the nation's recorded history. Despite the severity of the event, ground-based suppression systems exhibited critical limitations in early detection and timely response. While satellite-based approaches have emerged as effective alternatives, existing methods that rely heavily on brightness temperature thresholds and contextual comparisons often suffer from high false positive and false negative rates under varying environmental conditions and fire intensities. To address these challenges, this study developed a real-time wildfire detection algorithm using geostationary satellite imagery from GEO-KOMPSAT-2A and machine learning. The developed model was evaluated against existing wildfire detection products based on both polar-orbiting and geostationary satellites, using historic wildfire events that occurred in South Korea in March 2025. The model successfully detected all seven large wildfires that had failed initial suppression and achieved the highest overall performance, with a recall of 0.329, precision of 0.987, and F1-score of 0.494 across 79wildfire cases, including those with burn areas under 10 hectares. Moreover, the model provided the fastest early detection, with an average detection delay of only 12.9 minutes-significantly outperforming polar-orbiting satellites, which showed delays ranging from 197.2 to 305.2 minutes. By integrating geostationary satellites with machine learning, the model preserved the inherent advantages of geostationary platforms-such as continuous monitoring and early warning-while achieving detection sensitivity comparable to that of high-resolution polar-orbiting systems. These results demonstrate the potential of machine learning-based wildfire detection models to enhance the reliability and responsiveness of real-time wildfire monitoring and underscore the value of geostationary satellites in disaster management systems.