In this study, a comprehensive and cost-effective approach is proposed for classifying and analyzing various partial discharge (PD) scenarios in high-voltage power cables, aiming to enhance system efficiency. The methodology integrates signal processing techniques, extraction of physical-statistical features, and machine learning algorithms. To generate training data, detailed modeling of the cable structure was performed using COMSOL software, simulating diverse scenarios including: a healthy integrated cable, presence of a joint along the cable, single-source PDs (1 to 4 cavities), and combined cases involving joints with multiple cavity defects generating discharges. Subsequently, by analyzing the cable’s transient response under different discontinuities, the reflective signal patterns were characterized and fed into a machine learning model to identify the impact pattern of each scenario. The proposed model was then trained and evaluated in terms of classification accuracy, stability, and generalization capability. Overall, the results demonstrate that the combination of reflective signal analysis with time-frequency indicators and a support vector machine (SVM) algorithm provides a precise, robust, and interpretable framework for PD scenario classification. This method is suitable for deployment in power cable condition monitoring systems, even under noisy environments or limited data conditions.