A Domain-Aware Data Fusion Scheme for Cold Start Problem in Early-Stage IoT Systems

In recent years, edge computing has become increasingly popular, and edge intelligence is one of the key components of this paradigm. However, in the early phases of deployment, edge-based models frequently face an issue called the “cold start problem” caused by insufficient data. To address this problem, this work presents a data fusion model based on the ideas of domain adaptation. The model first evaluates the degree of similarity between the learning methodology’s source and target domains. Then, it appropriately combines data from both sources to train the model efficiently and reduce the possibility of overfitting. A lot of testing has been done to evaluate the effectiveness of this suggested learning model using the Ganga River and Sangam datasets. Compared to the Target and Source-only models, the findings show a notable improvement, especially in accuracy, during the initial training period.