The cost to create training data for supervised learning has been a problem. Particularly, it takes a long time to label time series data by hand. In this study, we propose a time-series training data creation support tool with macro and micro skip functions in order to improve the efficiency of manual labeling. The work process of the restaurant's customer service staff changes according to the time of occurrence. This is thought to be the cause of the covariate shift. The macro skip function preferentially selects the hour in which the covariate shift is considered to occur. On the other hand, micro skip function skips the time when occur probability of labeling target's service operation is low. Therefore, the micro skip function uses the estimated results from the classifier in progress of creation.