Four Uncertain Sampling Methods are Superior to Random Sampling Method in Classification

Active learning has been widely used because it can automatically select the unlabeled samples with the largest amount of information for manual labeling. Therefore, it could solve the problem of knowledge bottleneck. Selective sampling belongs to the active learning approach, reducing labeling costs to supplement training data by requiring labels to provide only the most informative, unlabeled examples. This additional information is added to an original, stochastically selected training set in the expectation of improving the performance of generalization of the learning machine[12]. Uncertainty sampling belonging to selective sampling is one of the key techniques of active learning, which uses a classifier to identify the least reliable unlabeled samples[1]. In addition, active learning is a method applied to classifier. In this paper, four methods under uncertainty sampling are used: Least Confidence Sampling, Margin of Confidence Sampling, Ratio of Confidence Sampling, and Entropy-based Sampling. According to the results, the four methods of uncertain sampling have higher accuracy than the random sampling method.