A novel approach to database selection using feedforward neural networks

Selecting an appropriate database is a common challenge for professionals, including web developers and machine learning engineers. Choosing the most suitable database for an application is important for maximizing its performance. However, because many features of different databases overlap, manually predicting the best database is difficult and prone to errors. To address this issue, a new approach is proposed using a Feedforward Neural Network (FFNN) for database selection. This method involves four steps: feature selection, dataset generation, neural network modeling, and database prediction. In the feature selection step, important features of seven major relational databases—MySQL, MS SQL Server, Oracle, IBM DB2, PostgreSQL, SQLite, and Microsoft Access—are gathered through web searches. These features are used to create a ground truth table. During dataset generation, 2,400 combinations of 75 features are generated, and labels for each instance are calculated using a weighted average method. The neural network modeling step involves selecting an optimal feedforward neural network based on its parameters and performance. The network is then trained using the Levenberg-Marquardt backpropagation algorithm. In testing, user input is provided (a selected set of features is fed into the pre-trained network), and the system predicts the best database with a mean squared error (MSE) of 5.16E-14.