PURPOSE: To more surely obtain a result of evaluation of time series data from a neural network by excluding data being unreliable for evaluating learning proress from an object of learning progress evaluation.

CONSTITUTION: By an output monitor part 21, an output of a neural network is monitored, and in the case the output is unnecessary for advancing progress of learning, its output is disregarded and excluded from an object of learning progress evaluation. This operation is executed by a learning object selecting means 22. As for a decision for determining whether it is necessary or not, there are the case when an operator executes it, and the case when a computer executes it automatically. In the case of the former, the operator looks at the output reflected on a CRT display being the output monitor part 21, and excludes it from an evaluation object by, for instance, a keyboard, etc., being the learning object selecting means 22. In the case of the latter, the output monitor part 21 collects all output data, and the learning object selecting means 22 consists of a structure provided with a reference value setting part for setting a reference value for excluding the data from the object of learning progress evaluation.