Predicting traffic accident severity is crucial for preventing accidents and ensuring the safety of vulnerable road users. This study presents a robust and intelligent framework for predicting Road Traffic Accident Severity (RTAS) using a deep learning model enhanced with bio-inspired optimization techniques. The method initiates with the collection of a comprehensive road traffic accident dataset containing detailed information on various accident-related factors. Data Pre-processing involves handling missing values and performing Exploratory Data Analysis (EDA) to ensure data quality and reveal hidden patterns. To refine input features, the Seagull Optimization Algorithm (SOA) is used for optimal feature selection, effectively reducing dimensionality while preserving critical information. The issue of class imbalance common in accident severity datasets is addressed using SMOTE (Synthetic Minority Over-sampling Technique) to synthetically balance the classes. The proposed framework is a novel deep architecture called Hierarchical Attention Wave Kernel Recurrent Neural Network (Hawk-RNN), which integrates temporal dynamics, hierarchical attention, and wave kernel transformations to model complex dependencies in the data. The model is evaluated using standard performance metrics such as accuracy, precision and recall, value of 92%, respectively. Experimental results demonstrate that the proposed approach significantly outperforms traditional models in accurately predicting accident severity, thereby offering a valuable tool for proactive road safety and traffic management.