Brain-computer interfaces offer transformative potential across a variety of fields, such as assistive technologies and neurorehabilitation. Traditional machine learning methods for P300 classification are typically subject-specific, which can lead to reduced generalizability. This study explores the subject-independent classification of P300 responses elicited through RSVP across 20 subjects. Three models — Bayesian Ridge, CNN-based, and EEGNet — were evaluated for their performance. The results revealed that EEGNet outperformed both Bayesian Ridge (ROC-AUC: 0.732) and CNN-based approaches (ROC-AUC: 0.763), attaining an average ROC-AUC score of 0.767. Additionally, the impact of varying the amount of training data was examined, demonstrating that larger training datasets significantly improved classification performance. Furthermore, fine-tuning EEGNet on individual test subjects significantly enhanced its performance, increasing the average ROC-AUC to 0.813. A paired t-test confirmed the statistical significance of the improvement, highlighting EEGNet’s robust potential for generalizable P300 classification.