Abstarct: Abstract In recent years, learning from data streams has attracted the attention of researchers and specialists. However, quantification learning has remained mostly unexplored. In some applications where we need to distribute positive and negative feedback, the use of quantification learning has been very useful. Also, this method can also be used to obtain specific general characteristics about the population of a network, and extract useful practical information by analyzing people's emotions. Quantification learning is very similar to classification, and both of them do the grouping of the data, but their purpose is different; In quantification learning problems, we are not looking for to specify each class of samples, and only general data statistics are important, and the goal is to provide an estimate of the distribution of data. Recent algorithms in the field of quantification learning in data stream have been introduced with the help of changing the concept and using the label request for a large part of the new samples and have been presented with sample selection techniques. In this research, the idea is to request the label of a smaller subset of recent examples and we make the classification model incrementally with the help of several different classification classes. Our experiments show that despite reducing the label request from recent samples and even removing it, the accuracy of the model can be maintained or improved.