Linear B-cell epitope prediction, as an in scilio evaluation tool in many immunological applications, has gained much attention in recent years. As epitope regions do not have a clear boundary, its prediction is generally difficult and inaccurate. The author proposes a hybrid model SSEPred that combines peptide sequence embeddings, multiple propensity scales and XGBoost, and conducts detailed research on the effects of peptide embeddings and propensity scales. The final model yields ROC AUC of 71.6% and F1-score of 80.9% by five-fold crossvalidation on reduced IEDB linear B-cell epitope dataset, which outperforms most existing methods.