Modern IoT networks base mission-critical event monitoring on dense wireless sensor deployments (IEEE 802.15.4/6TiSCH), with energy efficiency and detection reliability as the primary concerns. Traditional always-active sensing strategies, however, squander 42–58% of energy on duplicated operations, and static sleep schedules cannot compensate for dynamic event patterns, resulting in either excessive power consumption (≥3.8 mJ/node) or missed detections (up to 15% false negatives). To address this, we propose an Adaptive Node Selection and Scheduling (ANSS) that optimises sensor activity through intelligent scheduling and energy-aware decision-making. The framework leverages dynamic network conditions to efficiently allocate resources and improve overall system performance. To address this, we introduce an Adaptive Node Selection and Scheduling (ANSS) system that integrates federated learning-based event prediction (87.3% spatial accuracy) with QoS-aware dynamic sleep cycles (50-300 ms tunable intervals) and TOPSIS multi-criteria decision-making to activate only 18–32% of nodes per epoch. Deployed on a 150-node Zigbee testbed, ANSS delivers better performance, achieving a 39.2% energy saving (2.1 mJ/node versus 3.45 mJ baseline), 93.7±1.8% detection accuracy (a 5.3 percentage-point improvement over LEACH-C), and a 28.5% increase in network lifetime (first node death at 1,824 epochs). The solution preserves sub-100 ms latency (mean = 72.3 ms, σ = 9.4 ms) under 15 dB SNR while consuming only 5 KB of memory, making it applicable to resource-limited edge devices. These contributions set a new benchmark for energy–quality trade-offs in industrial IoT monitoring systems.