Abstract: A computer training device and method is provided for training a deep learning algorithm to classify incoming emails as belonging to one of multiple categories. The deep learning algorithm uses a loss function to avoid a low precision risk caused by a number of received emails for at least two of the multiple categories being imbalanced (e.g., by at least two orders of magnitude). The loss function compensates for the imbalance in received emails by changing depending on the criticality of the score being determined.