Parameter and Hyperparameter Optimisation of Deep Neural Network Model for Personalised Predictions of Asthma

Abstract

Over the last couple of decades, numerous optimisation algorithms have been introduced to optimise machine learning models. However, until now, no evidence or framework can be found in the literature that adequately describes how to select the best algorithm for parameter and hyperparameter optimisation of the Deep Neural Network (DNN) model. In this paper, an enhanced Fragmented Grid Search (FGS) method has been introduced for tuning several hyperparameters and finding the optimal architecture of the DNN model using less computation power and time. Furthermore, several experimental models are trained on the asthma dataset using various optimisers to find the optimal parameters, which can help the DNN model converge towards the lowest loss value. The results show that the Adam optimiser provides the best accuracy rate (96%). Consequently, the optimised DNN model can be used for accurately providing personalised predictions of asthma exacerbations for effective asthma self-management.