Financial statements are the reports prepared by organizations every quarter to show how the company has been doing since the last report and overall financial health of business. It usually contains various numerical data such as earnings, gross margin, profits, employee strength and so on, along with textual reports on the performance. These reports provide shareholders and general investors with critical information on the overall financial health and gives an indication on future roadmap. Due to the complex business processes, there are a lot of scope of including misleading information in these reports, as seen in major financial scandals over the past decades like Enron, General Electric, WorldCom, etc. This study collected financial reports for the US organizations from SEC in the year 2022 and used AAER reports in SEC to label fraudulent financial reports. The data is severely imbalanced as number of organizations appearing in AAER reports is insignificant as compared to total number of organizations filing reports in SEC. This study used a GAN-based model to generate synthetic fraudulent data to address the class imbalance problem and then applied ensemble classification models on the original and augmented dataset for training and classification of fraudulent statements. Finally, it compared the performance of the developed models against the recent available research results. It found the LightGBM model providing the best results in terms of prediction on test data, but surprisingly it performed better on original data than augmented data. As financial reporting fraud has proven to be the least frequent but most costly form of occupational fraud around the world, there is a lot of scope of future research for effective detection.