The problem of predicting sentiment from customers’ reviews has been an important issue for many years. Different machine learning methods have been utilized to sentiment polarity detection which identifies whether an input text is positive, negative or neutral. However, these methods suffer from low accuracy and recall. In order to build an accurate model for predicting sentiment polarity from social reviews, this paper presents an ensemble learning method-stacking generalization. The basic concept of stacked generalization is to fuse the first-level base classifiers’ outputs using a second-level meta classifier in a stacking manner. The diversity among the base classifiers with different features and weight measures is investigated in two domains (Twitter and Amazon product review), which gives a space for improving sentiment classification performance. Four types of base classifiers - support vector machine, boosted decision tree, bayes point machine, and averaged perceptron- are used to build the stacking model to use diverse features. The performance of singular and multiple base classifiers is compared with the proposed stacked ensemble learning model. Different evaluation measures, such as accuracy, precision, recall, and F1-score, are used to evaluate the proposed model’s efficiency. Results show that stacking has been proven to be consistently effective over both domains and working better than singular and multi-base base classifiers