How to use SPSS neural networks to classify data

The method of classifying data with SPSS neural network is as follows:

Neural network algorithms can gradually establish and improve the development path between input variables and output results through a large amount of historical data, that is, neural networks, in which the establishment of each nerve and the thickness (weight) of the nerves are trained by a large amount of historical data, and the more data, the closer the neural network is to reality. Once the neural network is established, it is able to output results with different input variable values. For example, a bank can build a neural network model based on historical customer information that applies for a loan, which can be used to make a decision on whether to lend to a customer in the future for a loan.

In this article, we will use a specific bank case to introduce how to use SPSS to build a neural network model to determine the repayment ability of future loan applicants.

When selecting historical data to build a model, historical data is generally divided into two parts: the training set and the validation set, and many analysts will directly use the first 70% of the data as the training set and the last 30% of the data as the validation set according to the data order. This is fine if the data can be shown to be independent of each other, but in the process of data collection, the collected data is often not completely independent (the correlation between variables may not be discovered by the analyst).

Therefore, the common practice is to use a random number generator to randomly divide the historical data into two parts, so as to avoid the data with the same attributes being classified into a dataset as much as possible, so that the established model can be more effective.

Before introducing the example of how to build a neural network model using SPSS software, let's introduce another function of SPSS: the random number generator. The random data of the random number generator constant of SPSS is not a true random number, but a pseudorandom number.

Pseudo-random numbers are calculated by an algorithm and are therefore possible. When the random seeds (algorithm parameters) are the same, the resulting set of random numbers is exactly the same for the same random function. The counterpart to pseudo-random numbers is true random numbers, which are true random numbers and cannot be ** and have no periodicity.

At present, most chip manufacturers have integrated hardware random number generators, for example, there is a thermal noise random number generator, which uses the thermal noise signal caused by the thermal vibration of the electrons in the conductor as a random number seed.

I can't just train, I do it with clementine, I do a lot of this kind of data analysis for others.