DETECTION AND ADAPTATION TO CONCEPT EVOLUTION IN DATA STREAMS: AN INTEGRAL FRAMEWORK

In the era of big data, machine learning systems face the challenge of adapting to dynamic environments, where data patterns change unpredictably, known as concept drift. In addition, new data classes may emerge, a phenomenon known as concept evolution, which represents a growing challenge in many real-world applications. Most current approaches focus on changes in data but lack efficient mechanisms to handle new classes. An additional problem is the availability of labeled data, as many algorithms assume that labels will be continuously available, which is unrealistic. Furthermore, many methods rely on user-defined parameters, which can affect performance. This paper proposes an integrated framework that combines data mining techniques to handle both concept drift and concept evolution in data streams, efficiently adjusting models to maintain performance in non-stationary environments. Results on two real-world datasets demonstrate the effectiveness of the proposed framework.