<p>Identifying influential nodes in complex networks is a fundamental problem with significant applications in areas such as information diffusion, epidemic control, and social network analysis. This research proposes a new deep learning-based framework, Inf-DLFS, which integrates deep feature engineering, optimization, and dynamic propagation modeling to accurately identify influential nodes. The method comprises three main components. First, a multistage artificial neural network (MS-ANN) is employed to extract informative node-level features by leveraging structural properties like degree, closeness, betweenness, and eigenvector centrality. This produces robust feature representations embedded with neighborhood context. Second, an improved selfish herd optimization (ISHO) algorithm is used for optimal feature selection. ISHO iteratively evaluates feature subsets using K-means clustering and silhouette scoring, thereby reducing dimensionality and selecting the most relevant features for influence prediction. Finally, the influence scores are computed using the impulsive pantograph neural network (IPNN), which models influence propagation through time with delayed feedback and periodic impulses. The proposed IPNN dynamically updates node states based on historical interactions and graph topology over multiple time steps. Extensive experiments on real-world datasets demonstrate that Inf-DLFS significantly outperforms recent deep learning methods in terms of accuracy, stability, and scalability. The integration of graph-based structural features, bio-inspired optimization, and dynamic modeling makes Inf-DLFS a powerful solution for influential node identification in complex networks.</p>