Proteins contain epitopes that serve as their antigenic determinants, with antigenicity referring to an epitope's ability to react with an antibody. These antigenic determinants exist as either continuous or discontinuous epitopes and are pivotal in integrative biology. The presence of B-cell epitopes has been extensively studied due to their potential in synthetic vaccine design. Thus, identifying antigenic sites on proteins is critical for developing synthetic peptide vaccines, immunodiagnostic tests, and antibody production. Recent years have seen the development of several computational methods, particularly machine learning-based approaches, to predict B-cell epitopes. However, these methods often fall short in reliably predicting linear B-cell epitopes, presenting an ongoing challenge. In this study, we propose and develop a novel method that integrates multiple deep learning models, aiming to harness the strengths of each model type while mitigating their individual weaknesses. During the model construction process, we addressed data imbalance and fine-tuned the model parameters to achieve optimal performance for our dataset. Our model demonstrated promising results, achieving an ACC of 0.846, AUC of 0.808, BAC of 0.716, MCC of 0.568 and PR-AUC of 0.705 on an independent dataset. This represents a significant performance improvement over previous algorithms.