Wireless Sensor Networks (WSNs) comprise energy-constrained sensor nodes that cooperatively monitor physical environments and forward sensed data to a base station. Prolonging network lifetime while maintaining reliable and low-latency communication remains a major challenge, particularly in cluster-based architectures where cluster head (CH) selection significantly influences energy consumption and transmission stability. Although numerous CH selection strategies have been proposed, many fail to simultaneously optimize energy balancing, delay reduction, and throughput enhancement under dynamic network conditions. To address these limitations, this study proposes a novel Energy-aware Multivariant Clustered Proportional Hazard Generative AI (EnMCPHG) framework that integrates deep learning-based energy modeling with statistical reliability analysis for efficient data transmission. The novelty of the proposed approach lies in the joint incorporation of a deep belief network for residual energy analysis, an energy-aware multivariant clustering mechanism to group nodes with similar energy profiles, and proportional hazard regression to predict the most reliable neighboring CH based on throughput performance. This hybrid generative and statistical framework enables adaptive CH selection and stable inter-cluster communication, thereby reducing transmission failures and latency. Simulation experiments conducted with 500 sensor nodes deployed over an 1100 m × 1100 m area using the Random Waypoint mobility model and Dynamic Source Routing protocol demonstrate the effectiveness of the proposed method. The results show a 31% reduction in packet drop rate, 25% decrease in end-to-end delay, 5% improvement in packet delivery ratio, 7% increase in throughput, and 22% reduction in overall energy consumption compared to conventional approaches, significantly enhancing network lifetime and overall performance.
