Domain-adaptation-based active ensemble learning for improving chemical sensor array performance

Once chemical sensors have been utilized for a long period of time, their sensing performance inevitably degrades, which means that the collected data drift over time, resulting in differential data distribution changes. Directly replacing sensors is laborious and costly. This research introduces a novel framework, named domainadaptation-based active ensemble learning (DAEL), that improves the gas recognition performance of a drifted chemical sensor array. The proposed framework utilizes source domain samples and a limited number of guidance samples from the target domain to develop a robust ensemble classifier. In this framework, two algorithms, called cross domain-adaptation-based active ensemble learning (DAEL-C) and discriminative domainadaptation-based active ensemble learning (DAEL-D), are proposed; these methods train ensemble classifier models for domain knowledge transfer by using different learning strategies for samples from the source domain and guidance samples from the target domain, respectively. Through an experimental validation conducted on a drifted sensor dataset, the proposed DAEL approach exhibits a more effective drift compensation mechanism than several existing comparative methods.