Introduction: Rheumatoid arthritis (RA) is an autoimmune disorder in which the immune system mounts an abnormal response to self-antigens, resulting in chronic inflammation and joint damage. Identifying antigenic regions in proteins that trigger RA is essential for the development of protein-based therapeutics. Methods: We developed predictive models for HLA class II binding RA-inducing peptides using a dataset of 291 experimentally validated RA-inducing peptides and 165 RA non-inducing peptides. Positional and compositional analyses were performed to identify residue preferences. Alignment-based approaches (BLAST and MERCI), machine learning classifiers, deep learning, and protein language model-based methods were evaluated for predictive performance. Results: Compositional analysis revealed significant enrichment of glycine, proline, and tyrosine in RA-inducing peptides. Alignment-based approaches provided high precision but limited coverage. Among machine learning methods, XGBoost achieved the best performance (AUC = 0.75) on the validation dataset, while ProtBERT was the top-performing protein language model (AUC = 0.72). The ensemble model integrating XGBoost with MERCI-derived motifs yielded the highest overall performance (AUC = 0.80; MCC = 0.45) on an independent validation dataset. Discussion: This study presents computational strategies for identifying RA-inducing peptides and demonstrates the advantage of combining motif-based and machine learning approaches for improved performance. The findings are valuable for evaluating the safety of proteins in probiotics, genetically modified foods, and protein-based therapeutics. To facilitate broader use, the best-performing approach has been implemented in RAIpred, a web server and standalone software tool for predicting and scanning RA-inducing peptides, available at https://webs.iiitd.edu.in/raghava/raipred/.