Sentiment Analysis is the study of determining an author’s opinion from written text using artificial intelligence and data mining techniques. In this thesis, three different sentiment analysis techniques; Naïve Bayes Classification, Support Vector Machine and Ensemble Classification are studied and applied to social media datasets for extracting opinions. One of the uses of sentiment analysis is to act as a feedback mechanism to aid in decision making. In this thesis a Probabilistic Feature Weighting (PFW) technique is proposed using the principle of the Naïve Bayes Classifier and Bayesian Probability. The PFW helps in ranking the documents into further sub-categories and is useful to compare features and their importance in sentiment classification. An Enhanced Ensemble Classifier Framework (EECF) is also developed based on the PFW technique. The Enhanced Ensemble Classifier increases the accuracy of the system compared to the existing techniques. Social media documents consist of a smaller number of words and often lack formal use of language. As such, social media requires more sophisticated techniques to establish sentiment. EECF helps in classifying shorter documents that have a smaller number of features such as Twitter posts. The development of the Enhanced Ensemble Classifier is a contribution in the sentiment analysis domain. The proposed PFW technique provides an alternative method to investigate features and classify sentiment into sub-categories beyond positive and negative sentiment. The Enhanced Ensemble Classifier that utilizes the PFW is shown to improve the determination of sentiment.