Selection an appropriate database is one of the problems faced by experts, ranging from web developers to machine learning engineers. Choosing the best database according to the requirement of the application can harness the full potential of the application and database. Since a large number of features of each database are overlapped with other databases, it is very difficult to predict the suitable one manually. Moreover, it is subject to human prone errors to provide accurate prediction. A novel database selection approach is proposed using Feedforward Neural Network (FFNN). This approach includes four steps: Feature selection, Dataset generation, Neural network modeling, and Database prediction. Feature selection involves searching the web to find the important features of the top seven relational databases selected for the prediction namely are: My SQL, MS SQL Server, Oracle, IBM DB2, Postgre SQL, SQLite, and Microsoft Access. Using the collected features, the ground truth table is created. Dataset generation includes the generation of 2400 combinations of 75 features and calculating labels for each instance using the weighted average method. The neural network modeling step consists of the selection of a feedforward neural network having the best parameters and performance. Finally, the neural network is trained using the Levenberg-Marquardt backpropagation algorithm. The system was tested with user input ( selected set of features fed to the pretrained network) and the best database was predicted with a minimum mean squared error (mse) of 5.16E-14.