A.S. Panfilova - Post-graduate Student, Moscow City University of Psychology and Education

Currently there is a tendency of influence of computerization on the development, production and performance tests, in addition to the established practice of using computers to count test results. New approach allows to work without a fixed number of questions, test continues until a decision support system gives a result about a certain level of intelligence abilities measured with the educational structure  Self-Organizing Feature Maps. The sample of the obtained values of the observed parameters for different ability levels is used for training the self-organizing feature maps of appropriate dimensionality. The aim is to obtain samples of Euclidean distances between an input vector, which describes a process of testing taken by a subject, and the trained network winning neurons - centers. Taking into consideration a sufficiently high dimensionality of input vectors, which is typical of practical problems, one can point out that the distribution of obtained Euclidean distances is near-normal. Sample estimates of mean distances and variances of these distances identify this distribution and allow estimating the probabilities of exceedence of distances between the obtained vector of subject-s responses, as well as the time spent for responding to every test task, and the corresponding center of the winning neuron, which makes it possible to judge the degree of model adequacy to observations. It allows to reduce testing time. Elimination of environment influence results is performed by comparing the observed and predicted answers to the test tasks using the Kalman filter, which is adapted to solve the problem. Kalman filter can process present-day data on the subject-s responses in real time, by forming its estimates immediately after obtaining the next response, without full test log, which is unavailable until the test task procedure is fully completed. Corrected response vector of the test according to each level of abilities is input to the corresponding trained self-organizing feature map, which can take a probabilistic assessment of testee belonging to one of the models. The presented decision support system can be used for various intelligence tests.