Metaheuristic optimization based feature subset selection with deep belief network for stress prediction in working employees

Stress is a physiological response to mental, emotional, or other physical challenges that humans face in real time situations, particularly in working environment. In recent times, there is a considerable increase in work related stress among working employees. The conventional stress assessment approaches are only based on self-reported questionnaires that are subjective and could not offer instant details regarding the person's condition. So, continual observation of stress using the recent technologies can be used to effectively comprehend the stress patterns and offer better understandings about probable future involvements. So, the advent of deep learning (DL) models assists the stress prediction process to understand the patterns effectively and offer effective perceptions about possible future interventions. In this view, this article presents a gravitational search algorithm (GSA) based feature selection with deep belief network (DBN) model, named GSAFS-DBN to predict stress among working employees. The proposed GSAFS-DBN model aims to predict the stress level by the selection of features and optimal classification process. In addition, the GSAFS-DBN technique involves the design of GSA based feature selection technique to choose an optimal subset of features. Moreover, DBN model is employed for the classification process to determine the proper class labels. Furthermore, Adamax optimization algorithm is applied to improve the training process of the DBN model. The effectiveness of the proposed model is examined using an own stress prediction dataset with 1197 samples of employees gathered from schools, banks, and universities and so forth. A detailed comparison study is implemented to highlight the enhanced predictive performance of the GSAFS-DBN approach in terms of different evaluation measures like sensitivity, specificity, precision, accuracy, F-score.