Clarity-Level Classification predicts the degree of clarity of a response to a query. It is essential to the advancement of many NLP activities, such as conversational AI, customer support automation, and instructional technology. However, it is challenging to assess answer clarity due to unclear wording, incomplete answers, and the contextual dependence between questions and answers. This paper describes our involvement in the shared work on Clarity Classification that SemEval2026 Task 6 created in order to address these issues. Using question-answer pair regression and classification, we suggested a transformer-based method. To train our model, we used a refined transformer model that included DeBERTa-v3-base. To address class imbalance, we used class-weighted loss functions and oversampling to implement class balancing. Results from experiments show that our suggested approach accomplished competitive performance.