Just-in-Time Adaptive Classifiers—Part I: Detecting Nonstationary Changes
IEEE Transactions on Neural Networks
Just-in-Time Adaptive Classifiers—Part II: Designing the Classifier
IEEE Transactions on Neural Networks
Incremental Learning of Concept Drift in Nonstationary Environments
IEEE Transactions on Neural Networks
Hi-index | 0.00 |
Aging of sensors, faults in the read-out electronics and environmental changes are some immediate examples of time variant mechanisms violating that stationarity hypothesis mostly assumed in the design of classification systems. Such changes, known in the related literature as concept drift, modify the probability density function of measurements, hence impairing the accuracy of the classifier. To cope with these mechanisms, active classifiers such as the Just-in-time adaptive ones, are needed to detect a change in stationarity and modify the classifier configuration accordingly to track the process evolution. At the same time, when the process is stationary, new available supervised information is integrated in the classifier to improve over time its classification accuracy. This paper introduces a JIT adaptive classifier based on support vector machines able to track changes in the process generating the data with computational complexity and memory requirements well below that of current JIT classifiers integrating k-nearest neighbor solutions.