Online Transfer Learning for RSV Case Detection

Transfer learning has become a pivotal technique in machine learning and hasproven to be effective in various real-world applications. However, utilizingthis technique for classification tasks with sequential data often faceschallenges, primarily attributed to the scarcity of class labels. To addressthis challenge, we introduce Multi-Source Adaptive Weighting (MSAW), an onlinemulti-source transfer learning method. MSAW integrates a dynamic weightingmechanism into an ensemble framework, enabling automatic adjustment of weightsbased on the relevance and contribution of each source (representing historicalknowledge) and target model (learning from newly acquired data). We demonstratethe effectiveness of MSAW by applying it to detect Respiratory Syncytial Viruscases within Emergency Department visits, utilizing multiple years ofelectronic health records from the University of Pittsburgh Medical Center. Ourmethod demonstrates performance improvements over many baselines, includingrefining pre-trained models with online learning as well as three staticweighting approaches, showing MSAW's capacity to integrate historical knowledgewith progressively accumulated new data. This study indicates the potential ofonline transfer learning in healthcare, particularly for developing machinelearning models that dynamically adapt to evolving situations where new data isincrementally accumulated.