Identify Arabic dialect using ensemble model

Abstract

This paper proposes a model that classifies the Arabic dialect of 26 different dialects from a given text. We used the dataset provided by MADAR Shared Task on Arabic Fine-Grained Dialect Identification, sub-task 1. 19 teams participated in this task, and the highest accuracy achieved by the winning team was 67.33%. Our model’s accuracy outperforms that by 0.785%. The proposed model consists of 3 classifiers and then ensembles the predictions from the classifiers to predict the result. We used both n-grams features and probability predictions from the 6-label dataset provided by the organizers. Our model achieves 68.15% accuracy and 68.01% macro F1-score.