Online Detection and Adaptation of Concept Drift in Streaming Data Classification.

The dynamism of our digital universe systems presents a key challenge for predictive analytics. Ensuring the model's ability to generalize beyond the training data is crucial for reliable predictions. Recent research has primarily focused on understanding and addressing the phenomenon of concept drift: the changes in the system that afect the model's accuracy. This paper aims to resolve this issue by introducing an online technique implemented using Light Gradient Boosting Machine (LGBM) classifier for concept drift detection and adaptation. Online LGBM is adjusted incrementally based on the most recent information, allowing it to adapt to changing patterns in the data stream. The proposed technique attempts to leverage incremental learning, ensemble learning, and the sliding window method to deal with concept drift. The experiments on Electricity, Spam and MixedAbrupt Drift datasets result in higher accuracy 86.77%, 97.67%, and 65.25% as compared to the ofine LGBM with lower accuracy 71.52%, 90.86%, and 50.01% respectively. The hyperparameters of LGBM are optimized using the Bayesian method. [ABSTRACT FROM AUTHOR]