This paper presents an advanced and highly relevant application of deep learning — specifically LSTM with attention and SHAP — for long-term pavement performance prediction. The integration of multi-year, multi-variable time-series data and attention-enhanced forecasting represents a substantial methodological contribution. The authors excel in preprocessing, segmentation, and modeling depth. With slight improvements in layout, introduction framing, and generalizability discussion, the paper could form the basis for a high-quality journal article. It is highly suitable for PFDM presentation and near-ready for journal-level submission.