Nowadays, artificial intelligence (AI) algorithms, grounded in data-driven methodologies, are increasingly deployed to automate and enhance decision-making processes across various domains. However, their use in safety-critical areas necessitates integrating human expertise to ensure decisions uphold principles of security, fairness, and privacy. Our research delves into unexplored territory by analyzing 72 public transportation accident datasets from Data.gov, uncovering key insights into risky driving behaviors crucial for traffic safety. By incorporating these insights into AI models, we enhance their capability to identify hazards and make informed decisions. Utilizing Decision Tree and XGBoost models, we develop state-specific accident judgment logic, subsequently verifying model robustness through automated methods. This novel approach not only showcases the potential of combining human knowledge with decision-making algorithms in improving traffic safety but also serves as a model for future development in critical domains. Our findings advocate for a balanced approach to AI system development, emphasizing the importance of aligning automated decision-making with fundamental safety and ethical standards, thereby contributing to safer, more reliable, and equitable technological advancements.