Sentiment analysis on Twitter provides organizations and persons with quick and effective instrument to observe the public's perceptions of them and their competition. A modest number of assessment datasets have been produced in recent years to check the efficiency of sentiment analysis algorithms on Twitter. Researchers offer a review of eight publicly accessible as well as manually annotated assessment datasets for analyzing Twitter sentiment in this research. As a result of this evaluation, we demonstrate that is a widespread weakness of many when using these datasets performing at sentiment analysis the objective (entity) level is indeed the absence of different sentiment classifications across tweets as well as the objects contained in them.[1], As an example all of that "I love my iPhone but I despise my iPad." Could be marked with a made-by-mixing classify however the object iPhone contained within this Twitter post should be annotated with just a label with an optimism. To get around this restriction and enhance existing assessment We have datasets that provide STS-Gold a novel assessment of datasets in which tweets or objects (entities) remain tagged separately hence might show alternative opinion labels. Though research furthermore compares the various datasets on multiple characteristics such as an entire quantity of posts as well as vocabulary size and sparsity.[2] In addition, look at pair by pair relationships between these variables and how they relate to sentiment classifier performance on various data. In this study we used five different classifiers and compared them and, in our experiment, we found that the bagging ensemble classifier performed best among them and have an accuracy level of 94.2% for the GASP dataset and 91.3% for the STS-Gold dataset.