OPTIMIZED ENSEMBLED CONCEPT DRIFT DETECTION

Machine learning is a branch of artificial intelligence that is crucial for predicting trends in various industries. In most real-world scenarios their underlying patterns constantly change over time, resulting in the degradation of model's performances. This is labeled as concept drift. The following article introduces an approach to detect drifts, named Optimized Ensembled Concept Drift Detection (OECDD). For the testing stage there was the usage of both artificially made and real datasets containing sudden, gradual and incremental drifts. By evaluating the proposed approach against the Naive Bayes classifier and a variant containing the classic drift detection detection method (DDM), it was possible to observe that OECDD achieved the highest accuracy, while presenting a significantly longer processing time - with RAM consumption rates being similar across all alternatives. Concluding, the developed method is suitable for more complex and less time demanding scenarios. Future work will explore optimization improvements and real-world applications.