Identifying high-utility sequential patterns that accurately reflect customer preferences and cross-channel dynamics is becoming increasingly difficult as omni-channel retail grows rapidly and generates diverse, ever-changing transactional data. Conventional sequential pattern mining and recommendation techniques struggle to adapt to the constantly evolving utility landscape and often neglect the complex higher-order interconnections across entities, users, and channels. This paper introduces an Adaptive Hypergraph-Attention BasketNet (AHAB-Net) designed to dynamically extract sequential patterns of significant usefulness for omnichannel retail recommendation systems, aiming to address existing constraints. The dynamic hypergraph in the suggested approach shows how people interact with each other in stores. In this graph, hyperedges represent multi-item baskets, contextual features, and behaviors that are distinct to each channel. To learn the importance weights of items and sequences based on channel influence, utility, and temporal relevance, an adaptive hypergraph attention method is described. BasketNet uses attention-driven aggregation and sequential encoding to discover high-utility patterns, enabling better adaptation to changing consumer behavior. The system is taught from start to finish and can be updated in small steps. It lets you make real-time suggestions in large retail settings. AHAB-Net achieves substantially higher utility gain, recommendation accuracy, and pattern relevance than deep learning baselines, traditional high-utility sequential pattern mining methods, and multi-channel retail datasets. The results reveal that the system can now better capture complex cross-channel relationships and is more flexible when concepts change. Testing of real-world omni-channel retail datasets shows that AHAB-Net consistently does better than the best models, such as regular high-utility sequential mining and deep sequential models. The proposed method results in enhancements of approximately 14–18% in high-utility pattern detection, 12–16% in recommendation accuracy, and 10–13% in utility-aware recall, while maintaining consistent performance amidst concept drift. In conclusion, AHAB-Net accurately captures complex cross-channel dynamics and changing customer preferences, which leads to better suggestions that are based on utility. The results show that it performs 15% better overall and is 20% more adaptable than current methods. This shows that it might be used in next-generation intelligent retail recommendation systems.