Dynamically updated ensemble-based machine learning for streaming data

Abstract

Aspects of the present disclosure provide systems, methods, and computer-readable storage media that support dynamically updated ensemble-based machine learning (ML) classification. An ensemble of ML classifiers may be created from a plurality of trained ML classifiers. These initial ML classifiers may be trained using labeled data to generate predictions based on input data. When an unlabeled data stream is received, the unlabeled data stream may be provided as input to the ensemble to generate predictions. After obtaining labels for the received data, the labels and the unlabeled data stream may be used to train new ML classifiers. The new ML classifiers may replace older ML classifiers in the ensemble. In this manner, the ensemble of ML classifiers is used to perform predictions on high volume streaming data while being dynamically updated with ML classifiers that have learned changes in statistical distribution across more recent input data.