On the complexity of characteristic function sets for correct diagnostic problem solving

There are discussed some abilities and tools to evaluate the quality of results of intelligent data analysis (IDA) for diagnostic type tasks. Reliability (non-falsifiability) of empirical dependencies formed in the process of machine learning (interpolation-extrapolation) by examples/precedents is explored by means of so-called Characteristic Functions (partially defined logical functions). The process of ChFs generation from training sample of examples is based on analysis of precedents description similarities formalized as binary algebraic operation. To provide a diagnostics correctness a method for evaluating a training sample representativeness is proposed. This method is based on diagnosis causality analysis provided by the used IDA tools. Some estimates of ChFs computational complexity are presented.