Text classification on large-scale real documents has become one of the most core problems in text mining. For English and other languages many text classification works have been done with high performance. However, Arabic language still needs more attention and research since it is highly rich and requires special processing. Existing Arabic text classification approaches use techniques such as feature selection, data representation, feature extraction and sequential algorithms. Few attempts were done to classify large-scale Arabic text document in a parallel manner. In our research, we propose a parallel classification approach based on the Naïve Bayes algorithm for large volume Arabic text using MapReduce with enhanced speedup and preserved accuracy. The experiments show that the parallel classification approach can process large volume of Arabic text efficiently on a MapReduce cluster and significantly improves speedup up to 12 times better than the sequential approach using the same classification algorithm. Also, classification results show that the proposed parallel classifier has preserved accuracy up to 97%.