Vasyayeva T. Neural networks and evolutionary methods of data analysis in expert systems for medical diagnostics

In this thesis an important scientific problem was solved. The problem of development of the methods of data mining for medical expert systems (ES) that is considered. The ES that has been developed allows to detect sudden infant death syndrome (SIDS) risk range with using methods, algorithms and software of information procession with using modern computer hardware. This allows to increase detecting of SIDS risk range reliability and to begin preventive actions earlier. For the first time an expert system devoted to SIDS risk range detecting has been designed, which allowed to increase reliability of SIDS risk range detecting and to detect the risk range even while the mother is pregnant or right after the child has been born. The method of informative risk factors selection for a genetic algorithm has received future development, by implementing fitness-function developed. This allowed to achieve a high performance of selection informative SIDS risk factors by regulating factors number and classification errors correlation. A tree-like encoding approach of chromosome encoding for a boolean function has been improved. This allowed to deduce classification trees for detecting SIDS risk range. An approach of encoding a chromosome as a tree, which embodies the boolean function in a disjunctive-normal form has been improved. This allowed to deduce classification rules for detecting SIDS risk range. For the first time a ternary logic in boolean calculations of classification tree was proposed. This allowed to conduct classification even when the risk factors are fuzzy.