This study makes use of supervised machine learning approach to achieve the objectives of feature-level opinion mining. Opinion summarization uses machines learning algorithms which can be broadly classified into three categories namely supervised, semi-supervised and unsupervised. Supervised machine learning algorithms require a labeled training data for constructing an inference function. This labeled training dataset is very time consuming to construct or acquire. Semi-supervised learning algorithms use a large unlabeled dataset along with a small labeled training dataset. The unsupervised machine learning algorithms do not require such labeled training datasets. Since the data given to the unsupervised learning algorithms is in unlabeled, it is difficult to evaluate the accuracy of the output produced.