Rule Induction by STRIM from the Decision Table with Missing and Contaminated Attribute Values

The statistical test rule induction method (STRIM) has been proposed as a method for effectively inducing if-then rules from a decision table. Its usefulness has been confirmed by a simulation experiment and comparison with conventional methods. However, real-world datasets often contain missing and contaminated values. This issue has been examined and addressed by various conventional methods. This paper also focuses on the problem of missing and contaminated values after specifying an observation system model for them. Experimental results show that STRIM is extremely robust for rule induction from such a decision table, even if many such values are contained in the datasets.