2020/11/05 by Lucas Baier, Vincent Kellner, Baier, Lucas +5 · 1 citation
Computer Science · Decision Sciences · Engineering · #Data Stream Mining Techniques #Advanced Bandit Algorithms Research #Smart Grid Energy Management
paper · pdf · doi:10.48550/arxiv.2011.02738
Machine learning models nowadays play a crucial role for many applications in\nbusiness and industry. However, models only start adding value as soon as they\nare deployed into production. One challenge of deployed models is the effect of\nchanging data over time, which is often described with the term concept drift.\nDue to their nature, concept drifts can severely affect the prediction\nperformance of a machine learning system. In this work, we analyze the effects\nof concept drift in the context of a real-world data set. For efficient concept\ndrift handling, we introduce the switching scheme which combines the two\nprinciples of retraining and updating of a machine learning model. Furthermore,\nwe systematically analyze existing regular adaptation as well as triggered\nadaptation strategies. The switching scheme is instantiated on New York City\ntaxi data, which is heavily influenced by changing demand patterns over time.\nWe can show that the switching scheme outperforms all other baselines and\ndelivers promising prediction results.\n