Hyperparameter Optimization for Multi-Layer Data Input Using Genetic Algorithm

In the training process of the deep learning algorithm, the hyperparameter optimization consumes the most time. This yields a problem that the development time of the model is extended. In previous studies, hyperparameters were changed without strict criteria. In contrast, in this study, the method using a genetic algorithm is proposed. Since the algorithm generates the fitness function through model's verification time and accuracy, the algorithm can find their optimal value. In addition, the number of populations of genetic algorithms is adjusted to reduce the time required for deep learning training process. In order to evaluate the performance of the proposed algorithm, the X, Y-axis vibration dataset is used to diagnose motor fault. As a result, it was confirmed that the proposed algorithm can obtain the proper verification time and accuracy when the hyperparameter optimization of the deep learning algorithm using the multi-layer is used as input data.