Arabic dialect identification presents an interesting problem in the field of NLP, with a noticeable gap in existing systems. To address this gap, we aimed to create a robust Arabic dialect classifier. We fine-tuned three deep learning models for classifying Arabic dialects, distinguishing between 19 distinct varieties. Our models were trained on data from the MIT dataset, supplemented with manually collected data from Bahrain and Tunisia. The achieved accuracy results were satisfactory, reaching up to 87% accuracy in the top 1 results.