<p>Considering recent updates to the European Commission’s Renewable Energy Directive, a further increase in renewable energy sources is predicted. Assuming the expected change in the energy generation scenario, combined with the varying loads of different consumers and nonlinear devices, the complexity of the system is expected to increase. This development will negatively impact the primary indices used by utilities to monitor distribution system performance, underscoring their role in enhancing power service quality and avoiding potential penalties. Consequently, voltage sags in distribution systems are important events to monitor, as they can occur multiple times per year and may disrupt industrial production and affect other segments of society. Therefore, to address the limitations posed by modern distribution systems, machine learning (ML) techniques are increasingly being explored for voltage sag characterization. However, the topic has not yet reached a point of maturity, with most existing approaches relying on complex signal processing, elaborated feature extraction, or computationally demanding architectures, often limiting their real-world applications. With oscillographic data stored in intelligent electronic devices (IEDs) installed in modern distribution systems, the characterization of voltage sags using such unconventional techniques becomes feasible. This research aims to develop a method for characterizing voltage sags using RMS voltage signals and machine learning, focusing on three common scenarios in distribution systems: transformer energizing, motor starting, and faults. The proposed methodology can distinguish the three events mentioned from those occurring during the normal operation of electrical networks, offering valuable support to distribution system operators. The proposed method is compared with similar approaches and validated using the IEEE 34-node test feeder network, under various operating conditions and fault scenarios. Based on the observed behavior, the proposed SVM-based method demonstrates strong performance and promising results, achieving high accuracy across various tests and noise levels, and consistently characterizing voltage sags correctly. Additionally, this work proposes a low-complexity methodology. Results show that, with appropriate hyperparameter tuning, the solution presented consistently achieves high accuracy, even considering an SNR of 20&#xa0;dB. The findings confirm that effective voltage sag characterization can be performed without advanced hardware or complex features, making the method directly applicable in practical distribution system monitoring.</p>