NIS-Apriori Algorithm with a Target Descriptor for Handling Rules Supported by Minor Instances

Abstract

For each implication $\tau : Condition\_part\Rightarrow Decision\_part$ defined in table data sets, we see $\tau$ is a rule if $\tau$ satisfies appropriate constraints, i.e., $support(\tau )\ge \alpha$ and $accuracy(\tau )\ge \beta$ for two threshold values $\alpha$ and $\beta$ ($0<\alpha , \beta \le 1$). If $\tau$ is a rule for relatively high $\alpha$, we say $\tau$ is supported by major instances. On the other hand, if $\tau$ is a rule for lower $\alpha$, we say $\tau$ is supported by minor instances. This paper focuses on rules supported by minor instances, and clarifies some problems. Then, the NIS-Apriori algorithm, which was proposed for handling rules supported by major instances from tables with information incompleteness, is extended to the NIS-Apriori algorithm with a target descriptor. The effectiveness of the new algorithm is examined by some experiments.