Hyperparameter prediction in ANN and RNN : a case study in hotel domain

Abstract

Deep Neural Networks are a subfield of the subsymbolic paradigm of Artificial Intelligence. Usually, when one wants to use artificial neural networks for a specific task, the neural network should be configured with optimal hyperparameters. To conduct such a task, the user should know about neural networks, as finding the optimal set of hyperparameters is time-consuming when manually configuring. This matter is prominent in data science because the field's requirements are directed toward deep learning and LLMs with autoML platforms. Automated hyperparameter prediction and optimization results in knowing about configuration and turning neural networks entirely or partially eliminated. This research starts its journey to achieve hyperparameter prediction in a naval way by considering the selection of independent and dependent features. Hyperparameter generation happens by using neural networks and training 34 predictors and classifiers. Mapping between different feature sets and the existing model configurations, achieved by a naval general frame, has been introduced, using natural language processing to arrange feature names in similar columns. This nominal feature set is so diverse in such a way that it has 2-3 100% identical feature names per 40 feature name rows. So, naval natural language processing encoding was introduced as a solution. When the complete feature set is inserted into the model, the set of classifiers and the predictors achieve the configuration in RNN and ANN—the solution given the name HPPGeneral model. Then, the resultant configurations are applied to select problems in the hotel domain to evaluate the approach. This research has delivered several naval research outputs, such as general frame and natural language encoding, besides a model that can be used for both ANN and RNN. Finally, the hyperparameter prediction achieved by this approach also gives results almost similar to those of manual hyperparameter prediction.