This　study　introduces　a　novel　driver　safety　evaluation　method　for　online　car-hailing　drivers,　focusing　on　dangerous　driving　behaviors.　By　analyzing　data　from　97　drivers　over　one　month,　including　the　frequencies　of　dangerous　behaviors　and　driving　distances,　we　develop　a　methodological　framework　integrating　Data　Envelopment　Analysis　(DEA)　and　the　Super-Efficiency　DEA　model.　This　framework　dynamically　links　dangerous　driving　behaviors　to　driving　risk,　calculates　driver　safety　efficiency,　and　ranks　driving　safety.　It　also　offers　behavioral　improvement　advice　to　poorly　performing　drivers　based　on　slack　variables.　Compared　to　traditional　models　such　as　EWM-TOPSIS　and　Critic-TOPSIS,　our　method　exhibits　superior　stability　and　provides　a　more　individualized　and　objective　assessment.　The　DEA　model　effectively　ranks　driver　safety　performance,　highlighting　its　advantages　over　TOPSIS　in　handling　multiple　inputs　and　outputs　without　predefined　weights.　This　research　offers　valuable　insights　for　online　car-hailing　companies　to　identify　high-risk　drivers　and　implement　targeted　safety　training　programs.