Arabic text is one of the main challenges for machine learning and sentiment analysis to this day. In this paper, we introduce an Arabic text classifier that predicts both polarity and sarcasm. Six different supervised machine learning classification algorithms were used and gauged on our Arabic classifiers: Logistic Regression (LR), Multinomial Naïve Bayes (NB), Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF) and K-Nearest Neighbor (KNN) along with different N-gram for tokenizing and TF-IDF for feature selection. SVM shows the best accuracy results (F1-score of 58.5%) when predicting polarity, while DT achieves the best accuracy results (F1-score of 64.4%) when predicting sarcasm. Previous sarcasm classification research achieved (F1-score of 46%) accuracy using BiLSTM on Arabic corpus.