Salp Swarm Algorithm for Drift Compensation in E-nose

E-nose technology relies on the proper functioning of sensors to identify and discriminate between different chemicals and odors. The long-term reliability of e-nose technology is hindered by the phenomenon of sensor drift. The effect of sensor drift is seen as a random and unpredictable shift in the data domain. This random shift in data deteriorates the performance of machine learning algorithms used in e-nose technology. Swarm intelligence based optimization has been successfully applied in different domains to deal with NP-hard optimization problems. In this paper, a swarm intelligence-based metaheuristic is proposed to deal with the sensors drift issue in e-nose technology. The proposed framework is validated using a benchmark dataset of sensor drift, and a significant improvement is observed in terms of the classification accuracy of different industrial gases. The proposed framework has the following benefits over conventional approaches: (i) there is no need for sensor re-calibration; (ii) there is no need for sensor replacement; (iii) there is no need for target domain data; and (iv) there is no need for domain transformation. Instead, the proposed work relies only on the source domain data and optimizes the feature space to deal with sensor drift. This makes the proposed framework more suitable for real applications of E-nose technology.