This paper addresses the problem of learning in dynamic environments where the distribution of training examples changes over time. The authors propose a method for detecting changes in the probability distribution of examples, which is crucial for maintaining accurate predictions. The method controls the online error-rate of the algorithm by monitoring the error rate as new examples are presented. When the distribution changes, the error rate increases, and the method defines warning and drift levels to identify these changes. The algorithm then learns a new model using only the examples after the warning level is reached. The method was tested on both artificial and real-world datasets using three learning algorithms: a perceptron, a neural network, and a decision tree. The results show that the method effectively detects drift and learns new concepts, and it is independent of the specific learning algorithm used. The paper also discusses related work and future directions in handling concept drift.This paper addresses the problem of learning in dynamic environments where the distribution of training examples changes over time. The authors propose a method for detecting changes in the probability distribution of examples, which is crucial for maintaining accurate predictions. The method controls the online error-rate of the algorithm by monitoring the error rate as new examples are presented. When the distribution changes, the error rate increases, and the method defines warning and drift levels to identify these changes. The algorithm then learns a new model using only the examples after the warning level is reached. The method was tested on both artificial and real-world datasets using three learning algorithms: a perceptron, a neural network, and a decision tree. The results show that the method effectively detects drift and learns new concepts, and it is independent of the specific learning algorithm used. The paper also discusses related work and future directions in handling concept drift.