Design of Experiment-based Configuration of Hyperparameters Of An Artificial Neural Network

This work presents an application of Design of Experiments (DOE) to the choice of best configuration of hyperparameters in a neural network. The example provided uses real data and shows how the use of DOE principles can increase accuracy and reduce the effort required when tuning complex machine learning algorithms. This strategy is particularly useful for practitioners who do not have any particular expertise in this field and can help them understand the relationships that exist between the different hyperparameters and some relevant metrics such as computational time and validation error.