Neural network, as the name suggests, is a statistical model that simulates animal neurons for data analysis, usually used to deal with nonlinear or complex data.

How to use SPSS to create neural networks?

The neural network models in SPSS include two types: multi-layer perceptron and radial basis function. Here, I will demonstrate the method of setting up a multi-layer perceptron model.

1. Import file. After entering the main interface window of SPSS, we switch to the "File" tab and use the "Open, New, or Import Data" command to import the raw data used to create the neural network.

Figure 1: Importing Files

2. Analysis - Neural Networks. Then switch to the 'Analysis' tab and click on the' Neural Network - Multilayer Perceptron 'option in its drop-down list.

Figure 2: Analysis - Neural Network

3. Set up a multi-layer perceptron. After entering the settings window of the multilayer perceptron, we drag the dependent variable, factor, and covariate to the corresponding target selection boxes according to the data type.

Figure 3: Setting up a multi-layer perceptron

4. Set up partitions. Then switch to the 'Partition' tab and set the partition ratio of the neural network according to the partition table, mainly including three items: 'Training, Testing, and Persistence'.

Figure 4: Setting up partitions

5. Set up training. Switching to the 'Training' tab, we can set the training type, including 'Batch, Online, and Small Batch'; The algorithm optimization methods below include [Scale Conjugate Gradient and Gradient Descent]. After completing the settings, click the 'OK' command at the bottom to start creating the neural network.

Figure 5: Setting up training

How to set up the hidden layer of SPSS neural network?

In the process of creating neural networks, setting the 'hidden layer' can avoid situations where the fitting state is too few or too many, simplify the number of neural layers, and create simpler data curves.

1. Architecture. In the process of creating neural networks, we can switch to the "Architecture" option, click on the "Architecture Auto Selection" option, and set the "Minimum Unit Number" of the hidden layer to 1 and the "Maximum Unit Number" to 50.

Figure 6: Architecture

2. Customized architecture. If you want to customize the architecture, you can set the hidden layers to one or two and activate the curve models below, including hyperbolic tangent curves and S-shaped curves.

Figure 7: Customized Architecture

3. Other settings. In addition, we can also set the number of units in the hidden layer, including automatic calculation and customization; The output layer below can also be used to activate special function formulas.

Figure 8: Other settings