The key worry for several sectors, including the government and consumers is financial fraud. Cloud computing and mobile computing have created more issues recently. Conventional manual detection methods take a lot of time, are inaccurate, and can't manage massive data on their own. Hence, a variety of methods have been used to address this extremely important issue of financial fraud. Instead of being produced by actual events, "Synthetic Financial Datasets for Fraud Detection" is synthetic data that has been created. Due to the confidentiality of financial services information, it was developed utilizing the mobile money payment simulator (PaySim). Customer and fraudulent behavior are present in the data produced by the simulator. The management of this data would be difficult because of its larger magnitude. This work has addressed different types of financial frauds involved during the transactions. The exploratory data analysis is applied to explore the features. Dataset is quite huge & unbalanced to process on conventional machines and therefore various sampling techniques were explored to balance the dataset for the better results in terms of accuracy and make the data set reliable. Dataset is divided into 15 chunks with 12 chunks for training and 3 chunks for testing purpose. Various classification techniques including ensemble techniques, Ada Boost, decision tree have been applied on each of the chunk. To ensure the reliability of the model, the results were compared with ensemble technique and decision tree classifier. With feature selection & dataset balancing, the model is showing 80 percent of accuracy.