<p>DNA sequence similarity analysis plays a crucial role in identifying evolutionary relationships and understanding unknown sequences by comparing them with known references. This study proposes an alignment-free approach to overcome the computational limitations of alignment-based methods, which struggle with large datasets and structural variations such as inversions and duplications. The proposed k-mer counting method (Algorithm-I) leverages multiprocessing to achieve efficient k-mer analysis with lower memory consumption and competitive execution times. For pairwise distance measurement, Algorithm-II introduces a local attention-based approach inspired by NLP techniques, improving phylogenetic tree accuracy and optimizing RF distance across multiple datasets. Evaluation with an MLP classifier demonstrates that the attention-based distance method achieves a training accuracy of 94.66% and a test accuracy of 89.41%, along with superior precision, recall, and F1-score.</p>