Identifying hope speech-text that conveys positivity and empathy online, is vital to building healthier digital communities. This study focuses on the English subset of the HopeEDI dataset, where hope speech accounts for only 8.6% of the data, posing a strong class imbalance challenge. To mitigate this, we apply a synonym replacement–based data augmentation technique and evaluate its effect across ten models, including five classical machine learning classifiers and five transformer-based architectures. Our results show that trans-formers, especially with data augmentation, substantially outperform traditional classifiers, achieving notable gains in recall and F1-score. These findings high-light that lightweight lexical augmentation can effectively enhance minority class recognition in limited datasets. The study contributes practical insights for developing inclusive, positive-content detection systems for online platforms.