Statistical Aspects of Classification in Drifting Populations
Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics, PMLR R1:521-528, 1997.
This paper discusses ideas for adaptive learning which can capture dynamic aspects of real-world datasets. Broadly, we explore two approaches. The first examines ways o f updating the classification rule as suggested by some monitoring process (similar to those used in a quality control problem), and this is applied to linear, logistic and quadratic discriminant. The second approach examines nonparametric classifiers based explicitly on the data and ways in which the data can be dynamically adapted to improve the performance. These methods are tried out on simulated data and real data from the credit industry.