Research Plans and Projects

Research interests include non-standard and robust stochastic models, classification, and pattern recognition. Areas of application include the physical and medical sciences and engineering.

Currently I have four major research projects underway in these areas.

The first project concerns robust pattern recognition. This studies the robustness of several discriminant analysis problems when the training data is contaminated with bad data points. A measure of this robustness called the "training data breakdown point' has been proposed by this investigator. This concept has been evaluated for several common discriminant procedures. The results have appeared in papers jointly written with my former Ph.D. student, J. Yang, and show that many popular discriminant procedures are very non-robust.

The second project, in the area of pattern recognition is to continue to develop the concept of "admissible clustering procedures". This concept was invented by Lloyd Fisher and myself in 1975 and remains today one of the few global objective criteria for evaluating clustering algorithms.

The third project concerns the measurement error model. The measurement error model is similar to the regression model only it assumes that the regressor variables (explanatory variables) are measured with error as well as the dependent variables. This model is probably more common in applications than the standard regression model though most users are unaware of the measurement error model and erroneously use standard regression. In particular, I am developing robust measurement error methods. I have co-authored several papers and a book with another former Ph.D. student, C.L. Cheng, on robust procedures and applications to the calibration problem. The book, is on the statistical properties of the many variations of the measurement error model including nonlinear models, robust models and Berkson models. This monograph is Volume 6 in the Kendall's Library of Statistics series published by Arnold Publishers, London.

The fourth project is a medical application concerned with prostate cancer. I have been involved since 1970 with various quantitative problems in the medical area of urology. The current project is to give some measures of the likelihood of finding a prostate cancer lesion using random prostate needle biopsies.

Visit Dr. Van Ness' WWW page