Authors are developing precedent approach to solving the problem of optimal decision making. The method they develop makes it possible to make the most adequate precedent selection in conditions where the object under consideration is not fully described, and cannot be estimated unambiguously. The originality of the approach offered by authors is in its focus on functioning with varying set of features (attributes). It is important for different applications, but it is especially important while supporting physician’s decision making, who often has a lack of time and resources. The method presumes the need in differentiating possible object membership that may be done by widening of its feature space. This task may in its turn be reduced to investigating of feature’s roles and their combinations (as in differential diagnosis and semiotics in medicine). In order to determine in what way should one retrieve missing features the authors offer to use the following conceptions: range, persistent feature combination, frequency of occurrence, availability of a feature, and object category.

. Decision support systems where the results of deduction on rules are supplemented with Case-based Reasoning results are another step forward in comparison with models which are supporting only single knowledge paradigm. The main deduction instruments in up-to-date hybrid systems are producing rules. The precedents are used only for exception processing. The approach described here which was designed and developed for the second release of system in the Institute for System Programming of Russian Academy of Sciences. On the basis of this research system the medical decision support system “Doctor’s Partner” is being developed. Now the rule and precedent reasoning are complementing each other while the conditions of ambiguity of not fully described case still exist. The absence of significant feature for one of the rules often prevents from using the rule-oriented method at all. During the current project stage the method of two-phase case estimation is used. The first phase consists of using the Case-based Reasoning method in order to get some knowledge about the possible case class (so called differential set). The next phase will use this information to produce the reverse logical inference treating the possible conclusion as a hypothesis and moving to the facts that are able to confirm it. The approach shortly described here may decrease the cost of additional rule investigation and is significantly limiting the branch choosing while moving along the rules chain.

The problem described concerns with multi-parametric control of a very complicated object as a process with complex interference of actions. For such objects it is hard or even impossible to obtain an adequate behavioral model. The approach to the “control process” formalization is developed. This approach is based on the theory of adaptive control, problem solving, data mining and case-based reasoning. The goal of programming system design is to provide a facility for adequate control under conditions of numerous external actions with complex non-parameterized interference.

Doctor's Partner is an intelligent system for physician's decisions support in Diagnostics and Choice of Treatment. Its approach is based on using the Case-Based Reasoning. A natural extension of its possibilities would be bringing in experience, accumulated by other physicians in the similar systems. The transition to the distributed version of the system is described along with the exchange between local case bases. This exchange, in essence, is a knowledge exchange. The options for such an exchange are described.

The integrated approach for building the physician's decision support systems was developed. This approach is based on using the Data Mining in Case-Based Reasoning. The technique was used which permits the lack of attributes for investigated object, so this object can be treated ambiguously. Using the approach, one can separate the case base into the equivalence classes in the attribute space and use the measure of closeness that takes into account the ambiguous estimations of the objects from the class intersections.

В статье предлагается подход к интеграции методов добычи данных, вывода на основе прецедентов и адаптивного управления в единой самообучающейся системе, позволяющей управлять объектами с плохо формализуемым поведением.