10K Financial reports are submitted by public listed companies to the Security Exchange Commission (SEC) yearly or quarterly. It allows investors to understand strategic planning and directions of the business organization. Although true facts are required to be presented in the reports, it does not prevent companies from using confusing explanations to beautify the organizations current state. Hence, an automated approach to filter out sentiments from the reports is crucial to assist investors in evaluating financial reports. This research paper explores machine learning approaches to conduct sentiment analysis on 10K financial reports. Two different datasets were intended to be used for training the model but only the financial phrase bank dataset was used to produce the final machine learning models. Four machine learning models including fastText, Naïve Bayes Support Vector Machine (NBSVM), Bidirectional Gated Recurrent Units (BiGRU), and Bidirectional Encoder Representations from Transformers (BERT) are trained based on the financial phrase bank dataset. It is discovered that the BERT model performed with the best accuracy while testing the models while the fastText model provided the fastest loading and training time. Conclusion of this research paper shows that different machine learning models in sentiment analysis possess respective advantages and disadvantages and further research can be done with the combination of textual and numerical data in financial reports.