Balancing Performance and Energy Consumption of Bagging Ensembles for the Classification of Data Streams in Edge Computing

In recent years, the Edge Computing (EC) paradigm has emerged as an enabling factor for developing technologies like Internet of Things (IoT) and 5G networks, bridging gap between Cloud services end-users, supporting low latency, mobility, location awareness to delay-sensitive applications. An increasing number solutions in EC have employed machine learning (ML) methods perform data classification other information processing tasks on continuous evolving streams. Usually, such cope with vast amounts that come streams while balancing energy consumption, predictive performance algorithms. Ensemble achieve remarkable when applied due several models possibility selective resets. This work investigates a strategy introduces short intervals defer mini-batches. Well balanced, our can improve (i.e., delay, throughput) reduce consumption bagging ensembles classify The experimental evaluation involved six state-of-art ensemble algorithms (OzaBag, OzaBag Adaptive Size Hoeffding Tree, Online Bagging ADWIN, Leveraging Bagging, RandomForest, Streaming Random Patches) applying five widely used benchmark datasets varied characteristics three computer platforms. As result, significantly 96% scenarios evaluated. Despite trade-offs, it is possible balance them avoid significant loss performance.