Photovoltaic (PV) systems play a growing role in the global shift toward renewable energy, but they remain vulnerable to a wide range of failures that can compromise performance and reliability. This study evaluates the effectiveness of the Performance Ratio (PR), a standardized metric from IEC 61724-1, as a tool for anomaly detection in PV systems. Using real-world data from five PV plants and manually labeled fault events, we assess the PR's ability to identify system anomalies and benchmark its performance against four machine learning (ML) models, including LSTM and Transformer architectures. Results show that the PR achieves an AUC score of 78.98, serving as a reliable baseline for fault detection, while ML-based virtual sensing models trained on meteorological and modeled irradiance data offer competitive accuracy. However, issues such as class imbalance and binary fault labeling highlight the need for more refined, granular classification methods. Ultimately, the findings support combining traditional PR metrics with advanced data-driven approaches to improve the robustness, interpretability and scalability of PV monitoring systems.