Neural Net Node

This information relates to a deprecated version of the Neural Net modeling node, and is provided here for reference only. A newer version of the node, with enhanced features, is available in this release. See the topic Neural networks for more information. Although you can still build and score a model with the deprecated version, we strongly recommend using the newer version.

The Neural Net node (formerly called "Train Net") is used to create and train a neural network.

Neural networks are simple models of the way the nervous system operates. The basic units are neurons, which are typically organized into layers, as shown in the following figure.

Figure 1. Structure of a neural network
Structure of a neural network

A neural network is a simplified model of the way the human brain processes information. It works by simulating a large number of interconnected processing units that resemble abstract versions of neurons.

The processing units are arranged in layers. There are typically three parts in a neural network: an input layer, with units representing the input fields; one or more hidden layers; and an output layer, with a unit or units representing the target field(s). The units are connected with varying connection strengths (or weights). Input data are presented to the first layer, and values are propagated from each neuron to every neuron in the next layer. Eventually, a result is delivered from the output layer.

The network learns by examining individual records, generating a prediction for each record, and making adjustments to the weights whenever it makes an incorrect prediction. This process is repeated many times, and the network continues to improve its predictions until one or more of the stopping criteria have been met.

Initially, all weights are random, and the answers that come out of the net are probably nonsensical. The network learns through training. Examples for which the output is known are repeatedly presented to the network, and the answers it gives are compared to the known outcomes. Information from this comparison is passed back through the network, gradually changing the weights. As training progresses, the network becomes increasingly accurate in replicating the known outcomes. Once trained, the network can be applied to future cases where the outcome is unknown.

Example. In screening agricultural development grants for possible cases of fraud, a neural network can be used for an in-depth exploration of deviations from the norm, highlighting those records that are abnormal and worthy of further investigation. You are particularly interested in grant applications that appear to claim too much (or too little) money for the type and size of farm.

Requirements. There are no restrictions on field types. Neural Net nodes can handle numeric, symbolic, or flag inputs and outputs. The Neural Net node expects one or more fields with the role Input and one or more fields with the role Target. Fields set to Both or None are ignored. Field types must be fully instantiated when the node is executed.

Strengths. Neural networks are powerful general function estimators. They usually perform prediction tasks at least as well as other techniques and sometimes perform significantly better. They also require minimal statistical or mathematical knowledge to train or apply. IBM® SPSS® Modeler incorporates several features to avoid some of the common pitfalls of neural networks, including sensitivity analysis (as indicated in the variable importance chart) to aid in interpretation of the network, pruning and validation to prevent overtraining, and dynamic networks to automatically find an appropriate network architecture.