Person identification using keystroke biometrics offers a scalable solution for identity verification in behavioral biometrics. This study introduces a framework where a transformer model serves as the baseline for capturing complex spatio/temporal patterns in keystroke features. To enhance accuracy, the model’s output is re-ranked using k-reciprocal nearest neighbors (k-RNN), which encodes neighborhood relationships into a feature vector for re-ranking under the Jaccard distance. The proposed method integrates misspellings and typos patterns, particularly using backspace key for correction, as a weighting factor to refine the final identification distance. This pipeline, combining the transformer baseline with k-RNN re-ranking and typing error adjustments, demonstrates significant improvements in person identification. Experimental results on Aalto keystroke database achive and EER about of 1.60. These findings validate the effectiveness of our proposed method and highlight its potential for secure and non-invasive applications.