To analyze gallup data multiple approaches were tried, but the following concept seemed to provide clearest results. Other attempts resulted in less definite group boundaries and/or larger variances in neurons. The phases are discussed in more detail in the case example.
The grouping of data can be performed visually, but it seems, that the human abilities make it a bit difficult to make clear decisions on which variables should be included in each group and which not. Data files, such as the one analyzed in the case example, usually contain lots of separate small groups. One good method is to use rules to verify, which variables dominate in each region. However, the regions should be formed at the same time as the variables are chosen. Therefore, some kind of optimization method is needed to form the groups of neurons indicating similar behavior.
A completely self-reliant process could be deviced to form the groups without human intervention. If there are lots of variables, all the possible combinations of variables cannot be verified. Therefore a genetic algorithm with random initial groups and a well constructed goodness criteria could produce good results quickly. However, the ideas for altering the population: refusal of entities, crossbreeding and mutations have to be considered carefully. There are also other choices of algorithms that can be used to find the optimal groups. This area would need some extensive research.