<p>Nocturnal predation remains a significant concern for the production of livestock, especially in rural and resource-constrained areas where standard remedies like physical barriers or manual observation frequently fail to provide adequate protection. In order to solve this recurring issue, we offer the Nocturnal Predator Detection and Alerting System (NPDAS), a real-time, edge-deployable artificial intelligence system designed specifically for infrared night vision enabled video streams. Unlike traditional thermal-based approaches, NPDAS only uses infrared cameras and leverages a lightweight yet powerful multi-stage machine learning pipeline optimized for autonomous predator identification along with early warning generation. The system uses Deep SORT for real-time object tracking after integrating a refined YOLOv8 model for initial object recognition in low light. A Bidirectional Long Short-Term Memory (BiLSTM) network is afterwards used to extract and classify temporal variables, such as dynamics of stride and movement patterns, to detect predator-like movements. A Support Vector Machine (SVM) classifier is employed concurrently to assess silhouette-based morphological characteristics to support classification determinations. A late fusion technique combines predictions from both classifiers, ultimately generating an alarm when predators are consistently identified across consecutive frames. Testing on a carefully selected night vision dataset shows that the system is highly effective, with 97.2% predator precision, 96.8% predator recall, and 98.1% livestock specificity. In addition, the enhanced precision and real-time capability of NPDAS are shown to be much higher than with other predator counters in the literature (YOLOv5, YOLOv7, Faster R-CNN, SSD, and RetinaNet). The proposed system can scale affordably to protect livestock in nighttime setting, supplementing animal welfare, mitigating financial loss, and bolstering the resilience of agricultural production from predators.</p>