Each individual has his own distinct character, making his own decisions which is based on his personality. Researchers in computer science field have tried to reach a model for extracting personality traits relying on user's profiles on social network sites as an input. Content created by users such as text posts, photos and shared activities in social network sites are considered as a huge source of data. Regarding user-created text, it has been proved that text pre-processing has a great impact if was applied to text before using it in research. In this paper, the effect of pre-processing (stemming and stop word removal) and adding numerical features is tested on the performance of Arabic personality prediction using AraPersonality dataset, which yielded 3.0% and 6.7% overall improvement to baseline experiments in binary representation and multiclass representation respectively.