A country�s federal government receives revenue from several sources. Example in Malaysia the sources are direct tax, indirect tax and non-tax revenue. The federal government will then use the revenue for operations and developments in the country. There are currently limited methods to predict federal government revenue for upcoming years. Having different and better method can help to better plan the collection activities and managing the resources. For now, Malaysia federal government can only forecast or estimate the revenue. Business intelligence on the other hand is currently booming in the business world as it helps to improve and provides relevant information for decision making process. One of the branches of business intelligence is predictive analytics, where it can be used to predict future outcomes provided past data are available. Patterns can be identified to predict the upcoming trend. From the observation, predictive analytics can be applied in any financial prediction which includes federal government revenue. Numerous machine learning methods exist such as linear regression, polynomial regression, various types of neural network, decision tree, random forest, multiple linear regression and so on. Based on the literature review done, feed forward neural network is highly used and thus selected for this study. Hyperparameter tuning is conducted to determine the ideal parameters for feed forward neural network to be applied for federal government revenue prediction. From the result, it is found out that using Softsign activation function and Adam optimizer can give better accuracy. Completing the study, it contributes to provide another way to accurately predict the federal government's revenue and subsequently be advantageous to the federal government. © 2023 Little Lion Scientific.