Abstract
In every text, words have various frequencies and keywords have strong relationship with the subjects of their texts. Word frequencies change due to time-series variation over given periods of time. An early method estimated stability classes that indicate word popularity due to time-series variation based on frequency changes in text data over given periods using a decision tree. The estimation precision of the decision tree decreases when there is scattering of data number among classes. This paper suggests a new way to use a Random Sampling Method and proposes a new Data Copying Method to improve the estimation precision of decision tree. By using this new Data Copying Method F-measures have improved: Increasing Class 9%; Relatively Constant Class 9%; Decreasing Class 18%.