In today's digital age, social media platforms have become a prevalent medium for individuals to share their opinions, emotions, and experiences. Despite the surge in user-generated content, effective tools and resources for sentiment analysis in the Arabic language remain insufficient. This paper addresses this gap by presenting a novel approach to Arabic sentiment analysis through the development of a web application for text emotion classification. The proposed methodology employs an ensemble of deep learning models, including Bidirectional Long Short-Term Memory (Bi-LSTM), Bidirectional Gated Recurrent Unit (Bi-GRU), and the MARBERTv2 transformer model, combined using a Random Forest stacking technique. The system's performance is evaluated on the Emotone_ar dataset, providing a robust benchmark for emotion detection tasks. Experimental results demonstrate that the ensemble model outperforms individual models, achieving an accuracy of 90%, a recall of 90%, and an F1 score of 90%. The integration of MARBERTv2, a pre-trained language model specifically designed for Arabic, shows superior performance compared to other models tailored for the Arabic language. This work concludes that the proposed ensemble model not only advances the field of Arabic sentiment analysis but also offers an effective tool for real-time emotion detection in social media texts, addressing a critical need in natural language processing for Arabic.