This is a deep learning model research into enhancing human DNA sequenceDNA sequences classification tasks task whose main focus has been on the pursuit of optimal architecture and feature representation. The paper points out outstanding performance achieved through the use of transformer-XLTransformer-XL against traditional machine learningMachine learning models such as logistic regression, Naïve Bayes, and random forestRandom Forest, reaching a record accuracy of 100% for the very first time. Transformer-XLTransformer-XL may be considered better for representing the long-term dependency and complex pattern in genomic big data, since the hybrid architecture, having LSTM + CNN, has only 83.22% as maximum accuracy. Other important aspects that this research has been able to study are the pre-processing and structuring of DNA sequenceDNA sequences data for deep learning models, reflecting the call for more studies to work on different encoding strategies toward better performance. Although extensive hyperparameter tuning was not a point of consideration, the results indeed tend to indicate that systematic optimization of parameters could result in further improvement in model performance. These results contribute to an accumulation of studies focused on deep learning in genomics and provide insight into possible future research targeted at optimizing deep learning models for DNA sequenceDNA sequences classification with a view toward practical applications.