Chong Ho Yu
Professor, Azusa Pacific University
Chong Ho Yu has a Ph.D. in Measurement, Statistics, and Methodological Studies and a Ph.D. in Philosophy of Science (Arizona State University). Currently he is a Professor of Behavior and Applied Science and also an Adjunct Professor of Mathematics at Azusa Pacific University. He is also the Quantitative research Consultant and the Committee Chair of Data Science Consortium and Big Data Discovery Summit at the same institution
WATCH LIVE: November 3rd at 10:00 am
Today many data science methods are highly automated and as a result the analytical process becomes a “black box.”
For example, the hidden layer in neural networks, as the name implies, is invisible to the data analyst. In rapid predictive modeling the result of the ensemble is selected by the algorithm. While pattern seeking is arguably the most important goal of data science, one potential pitfall of the “black box” approach is that the analyst could no longer visualize the data pattern and explore alternatives. In this presentation various data visualization techniques, such as the loading plot, profiler, multi-dimensional scaling plot, diamond plot, diffogram, bubble plot, and median-smoothing, will be illustrated. It is important to point out that with an extremely large sample size the analytical algorithm could be fooled, resulting in misidentify random noise as a systematic pattern. To rectify the situation, it is advisable to employ both data visualization and data science algorithms side by side.