THE RELATIONSHIP BETWEEN ERRORS AND SIZE IN KNOWLEDGE-BASED SYSTEMS

Previous researchers in knowledge-based systems verification have conc entrated on developing various approaches and computational tools to f ind errors in knowledge bases. However, unlike software engineering fo r traditional systems, there has been little investigation of the rela tionship between errors and system size. In addition, there has been l ittle analysis of the relationship between the occurrence of different types of errors. Thus, this paper investigates the empirical relation ships between knowledge-based system size and number of errors, and be tween the number and existence of different kinds of errors. It is fou nd that, in general, system size is statistically significantly correl ated with two of those error types, and with total errors. Further it is found that the size of ''smaller'' systems is not correlated with t otal number of errors, but the size of ''larger'' systems is correlate d with total number of errors. As a result, this evidence indicates th at it can be important to use a modular approach in the development of such systems. In addition, it is found that the number of different t ypes of errors have a statistically significant correlations with each other. Further, the existence of different errors types are statistic ally related. As a result, errors signal the existence of other errors . (C) 1996 Academic Press Limited