If-Then Rule Induction from Historical Data about Automatic Ticket Gate Machines

Abstract

Toward the discovery of useful information and knowledge from stored data, techniques for data mining and knowledge discovery have been studied remarkably. In this paper, a rough set-based approach to rule induction is applied to real world historical data about automatic ticket gate machines, in order to mine rules useful for the judgment of necessity of maintenance of the machine. By the rough set-based rule induction method, we obtain simple if-then rules. However, it is not very easy to induce if- then rules applicable for various homogeneous data from a given training data because there are too many condition attributes. Then we transform the daily data to cumulative data, i.e., accumulated daily data just after the maintenance day, considering the monotonicity between condition and decision attributes. We evaluate the improvement of induced if-then rules by the transformation by comparison with if-then rules induced directly from daily data.