Active Machine Learning in Regression Problems

In the scale of the big data, it is hard to track exact features from which machine learning algorithm has made associations. Moreover, it has no understanding of underlying processes, so it can be overly sensitive to changes in the training data. But as the data amount and complexity of machine learning algorithms are continuously growing, it is crucial to find mechanisms to ascertain their reliability and to reduce the amount of necessary data labeling to train the algorithm. Active machine learning may be one of the solutions. Its main objective is to achieve higher algorithm accuracy with less labeled data for training. But the mechanics by which the active learning algorithm selects the next data sample for labeling might be utilized to gain an understanding of associations created by the machine learning algorithm. This paper describes the development of the active machine learning method for regression problems. It results in a demonstration of the system's prototype which is used to train meta models for computer experiments.