Hybrid renewable energy systems (HRES) combining photovoltaic generation, battery storage, and backup sources are increasingly used in microgrids and remote installations. Despite improved sustainability, these systems remain vulnerable to short-term supply interruptions caused by irradiance variability, load changes, battery depletion, or backup unavailability. Conventional threshold-based alarms often fail to capture the multi-factor nature of outage formation, motivating data-driven prediction approaches. This paper presents an IoT-based monitoring and supervised learning framework for near-term outage-risk prediction in HRES. An Arduino-based edge device collects key electrical measurements and transmits time-stamped telemetry to a cloud platform via GSM/GPRS. Using a power-balance perspective, features are engineered to represent system adequacy, including irradiance conditions, battery state-of-charge (SoC), depletion trends, and backup availability. Horizon-based labeling is applied so that samples are marked as outage-risk when a failure occurs within a defined future window, enabling early warning rather than post-event detection. Three supervised classifiers-logistic regression, decision tree, and random forest-are trained and evaluated on a synthetic one-year dataset sampled at five-minute intervals. Performance is assessed using accuracy, precision, recall, F1-score, and ROC-AUC metrics. Results show that tree-based models, especially random forests, provide stronger recall and risk ranking performance, enabling earlier detection of outage-prone conditions. The proposed approach demonstrates that combining IoT telemetry with probabilistic prediction can support proactive actions such as generator start, load shedding, and maintenance planning, while providing a practical foundation for future validation using real operational data.