UMAP Based Data Validity Evaluation for Artificial Intelligence Systems

Although comparison is one of the most useful data validity evaluation method, it has a few potential drawbacks. One of them is that results from the training or the testing a machine learning model shall be obtained. Another one is that uncertainty due to the machine learning model itself may cause a difficulty in evaluating the data validity. In this paper, a new data validity evaluation method was proposed so that these drawbacks can be made up by the proposed method. The proposed method enables a data validity evaluation to be performed in data collection phase or data pre-processing phase. Uniform Manifold Approximation and Projection (UMAP) offers the methodological background to the proposed method.