Improving the Output Quality of Official Statistics Based on Machine Learning Algorithms

Abstract National statistical institutes currently investigate how to improve the output quality of official statistics based on machine learning algorithms. A key issue is concept drift, that is, when joint distribution independent variables and a dependent (categorical) variable changes over time. Under model requires regular updating prevent it from becoming biased. However, asks for additional data, which are not always available. An alternative reduce bias by means correction methods. In article, we focus estimating proportion (base rate) category interest compare two popular methods: misclassification estimator calibration estimator. For prior probability shift (a specific type drift), methods analytically as well numerically. Our analytical results expressions variance both As numerical result, present decision boundary relative performance provide better understanding effect quality. Consequently, may recommend novel approach use algorithms in context statistics.