PKDA model is based on kernel machines that exploit Bayesian derivation of classification problem using different kernel functions with model selection criterion based on the information-theoretic measure of complexity (ICOMP) index of this author which allows robust statistical inference in the kernel feature space.
An experimental case study on a real data set consisting of two breast cancer groups (“Benign”/”Malignant”) which is composed by n = 1269 Italian patients on p = 132 continuous features has been analyzed in detail. The efficiency and robustness of our approach is presented and compared as an alternative tool to the usual Support Vector Machines (SVMs) used in Computer System Detection (CAD) of breast tumors. According to our criterion, the best kernelized mapping is with the ERBF kernel function with a tuning parameter c = 2.602 which corresponds also to the lowest validation error rate of 37.97% with a gain of 14.64% over the usual Fisher Discriminant Analysis (FDA).
Such a result elucidates the current inferential problems in the classical statistical data mining and it
shows that our approach is a first step toward the specification of a robust classification model for breast cancer detection.
Dr. Hamparsum ("Ham") Bozdogan is Toby and Brenda McKenzie Chair Professor in Business, Information Complexity and in Model Selection in the Department of Statistics, Operations, and Management Science in the College of Business Administration at the University of Tennessee Knoxville (UTK), Tennessee. He is also a faculty member in the Center for Intelligent Systems and Machine Learning (CISML) at UTK.
Dr. Bozdogan received his B.S. degree in Mathematics, 1970 from the University of Wisconsin-Madison, and both of his M.S. and Ph.D. degrees in Mathematics, 1978 and 1981, respectively, from the University of Illinois at Chicago majoring in Probability and Statistics (Multivariate Statistical Analysis and Model Selection) with a full-minor in Operations Research. He joined the faculty of UT in the Fall of 1990. Prior to coming to UT he was on the faculty of the University of Virginia in the Department of Mathematics, and was a Visiting Associate Professor and Research Fellow at the prestigious “Akaike’s Institute,” The Institute of Statistical Mathematics in Tokyo, Japan during 1988.
HEALTHCARE SYSTEMS ENGINEERING - PAST, PRESENT, FUTURE
James C. Benneyan, Ph.D.
New England Healthcare Systems Engineering Partnership
NSF Center for Health Organizational Transformation
Institute for Healthcare Improvement, Faculty
Industrial Engineering and Operations Research
Problems with our healthcare system are well-known and staggering, including poor access, inefficient and ineffective processes, equity disparities, practice variability, and patient safety issues, all at enormous costs. An estimated $2.3 trillion annually continues increasing at almost double inflation, with ~30% attributable to poor processes, error, and waste. Estimates of medical errors include 1.4 million affected patients, 98,000 deaths, and $8.8 billion annually, while avoidable readmissions and patient non-compliance cost almost $200 billion/year. The enormity of such figures prompted the National Academy of Engineering, Institute of Medicine, and others to advocate greater application of systems engineering over a decade ago, yet not much has changed. Industrial and systems engineering, by whatever name, in fact has a long healthcare history, recently enjoying its fourth renaissance within academia. This talk is divided roughly into thirds - discussing this history, the present healthcare and IEOR landscape, and important future directions if our field is to have more profound impact.
Dr. James Benneyan is an industrial engineering professor at Northeastern University and director or co-director of NU’s healthcare systems engineering program, the VA’s New England Healthcare Engineering Partnership, and NSF’s Center for Health Organizational Transformation. His research spans healthcare systems engineering broadly – including probabilistic optimization, quality engineering, patient and drug safety, scheduling and logistics, risk-adjusted surveillance, and comparative effectiveness methods, with primary funding from NSF, NIH, AHRQ, NIDA, UNOS, and USAF. Benneyan has published over 75 papers in these areas; received 7 teaching, service, and research awards; and taught systems and design engineering to ages 6 through 60. He is past vice president of IIE, past president of the Society for Health Systems, a fellow of SHS, HIMSS, and IHI, and serves on numerous editorial boards. Prior to joining NU, Jim was senior systems engineer at Harvard Community Health Plan, industrial engineer at IBM and Digital Equipment Corporation, and statistical consultant at Productivity Sciences Incorporated.