<p>Road accidents are a significant public safety concern in Bihar, which demands effective predictive models to mitigate risks and enhance traffic management. Traditional methods for accident prediction often fall short owing to the complexity and nonlinearity of contributing factors, such as road conditions, traffic density, and weather. In this study, an artificial intelligence-based approach is proposed to predict road accidents using a Multi-Layer Perceptron (MLP) model optimized with three meta-heuristic algorithms: Bonobo Optimizer (BO), Smell Agent Optimization (SAO), and Dynamic Control Cuckoo Search (DCCS). The MLP model was trained on a dataset encompassing diverse features influencing road accidents and fine-tuned using metaheuristic algorithms to achieve optimal performance. The effectiveness of the proposed framework was assessed using evaluation metrics, such as R², RMSE, MSE, MDAPE, and WAPE. The results indicate that the hybrid MLP-meta-heuristic approach significantly improves prediction accuracy and provides actionable insights into accident causation patterns. This study underscores the potential of AI-driven methods as cost-effective and reliable alternatives to traditional predictive models, contributing to enhanced road-safety strategies in Bihar.</p>