TSU has taught neural networks to determine risk groups by psychotype
12.10.2016

The Department of Programming of the TSU Faculty of Applied Mathematics and Cybernetics has established a programme complex of several neural networks. It is able to solve one of the most important tasks of psychodiagnostics: to divide respondents into classes according to their psychotype. This also allows forming risk groups and predicting the behavior of an individual.

The neural network training was based on the results of psychodiagnostic tests of prisoners conducted in one of the Tomsk colonies by Leary methodology. This methodology reveals the dominant type of attitude of a person toward others. The materials analyzed were 180 questionnaires from those convicted under six statutes.

- We are interested in predicting behavior. Behavior is primarily determined by the personality and this temperament, character, and psychotype. The task for neural networks is to define a specific set of characteristics that distinguish those convicted on various counts, because it is a problem of classification - said Mikhail Golovchiner, project leader, Associate Professor at the Faculty of Applied Mathematics and Cybernetics.

The statute under which prisoners were convicted - murder, grievous bodily harm, theft, illegal possession and manufacture of drugs, and others - became the basis for determining the classification. The objective of the programme complex was to find the presence or absence of a certain combination of personality traits (dominant characteristics) of those convicted of each type of crime.

Two types of neural networks, a perceptron and a counterpropagation network (CPN), were implemented in the complex. The problem of classification was solved with the perceptron and then the results were confirmed or disproved by connecting the CPN modules.

- After training and testing, the complex revealed two characterizing factors, domination-subjection and friendliness-aggressiveness (hostility), and the classification accuracy was about 20%, indicating a relatively high accuracy, - said Mikhail Golovchiner.

Thus the perceptron has identified among those convicted for theft such dominant features as authoritarianism, dependence, and altruism, and suspiciousness among those convicted for the intentional infliction of grievous bodily harm.

The project manager emphasized that the programme complex could solve the classification problem for any group. He explained that the complex can be used for interviews in recruitment, if there are psychological portraits of the ideal and the unsuitable employee that could be the basis for classifying respondents.