COMPUTER INTENSIVE AND ROBUST METHODS FOR DATA ANALYSIS
Main researcher
: LUŽAR, VESNA (26962) Assistants
DOBRIĆ, VESNA (10093)
ŠIMIĆ, DIANA (125756)
Type of research: developmental Duration from: 01/01/91. to 12/31/93. Papers on project (total): 34
Papers on project quoted in Current Contents: 1
Institution name: Institut za medicinska istraživanja i medicinu rada, Zagreb (22) Department/Institute: Department for Applied Mathematics Address: Zagreb, Unska 3 City: 10000 - Zagreb, Croatia
Communication
Phone: 385 (41) 510-099
Phone: 385 (41) 278-963
Fax: 385 (41) 278-908
E-mail: vluzar@public.srce.hr
Summary: Because assumptions underlying standard methods for data
analysis are in many research studies not satisfied, development of robust
methods, less sensitive to departures from these assumptions, are of
considerable importance for multivariate data analysis and statistics.
Today's computing power allows for development and application of special
computationally intensive methods, particularly useful where classical
methods are unavailable or intractable. In this project, a class of robust
methods has been developed. For some of the methods, comparison studies
with the classical models have been performed. Also, three new distance
measures have been proposed. The problem of estimating biases of the
ordered characteristic roots of a random covariance matrix is reduced to
the computation of a definite multidimensional integral. Thus, it has been
approached by different simulation and variance reduction techniques. In
particular, the importance sampling method has been applied. Different
methods have been compared in terms of efficiency, defined as variance
ratio. The two-dimensional symetric interval eigenvalue problem has been
solved, and the results were used to define interval Jacobi method.
Numerical stability of the method was investigated. In the study of
Malecot's isolation by distance model, a novel table of 'critical' values
of Malecot's parameter "b" was introduced, and a simulation study for
analysing the power of the quadratic assignment procedure was designed and
developed. Computationally intensive and other methods for data analysis
have been applied in (1) industrial statistics: for predicting performance
of a aeroengine, for analysing traffic, car kinematics and pollution in
urban areas, in time series analysis of air quality data; (2) in citation
history analysis; (3) in biostatistics: for studying middle Dalmatian
population system by Malecot's isolation by distance model, and for
histological reclassification of chronic viral hepatitis.
Research goals: Main objectives of the project are: (1) Development
of algorithms for data analysis, that are robust to the departures from
the assumptions set by the classical models; (2) Comparison of proposed
models with the classical; (3) Development and application of simulation
and related methods, such as the importance sampling method for the
computation of multidimensional statistical integrals; (4) Development of
exact bounds for eigenvalues of a two-dimensional interval symmetric
matrix; (5) Numerical investigation on real matrices of numerical
stability of the Jacobi method; (6) Application of the above models and
other methods for data analysis and statistics to various problems in
industry (quality improvement), biomedicine and science, in general; (7)
To improve the collaboration with engineers, researchers, business
managers, medical and other professionals, in performing research studies,
applying sound statistical principles, appropriate statistical methods in
conjunction with their new responsibilities as managers of an emerging
market society; (8) Software development for (1)-(6).
COOPERATION - INSTITUTIONS
Name of institution
: Istituto Motori, CNR Type of institution: State institute Type of cooperation: Joint project City: 80125 - Napoli, Italy
Name of institution
: Universita degli Studi di Venezia, Facolta
di Economia e Commercio, Laboratorio di Statistica Type of institution: University/Faculty Type of cooperation: Occasional exchange of experts City: 30123 - Venezia, Italy
Name of institution
: The American University, Department of
Mathematics and Statistics Type of institution: University/Faculty Type of cooperation: Occasional exchange of experts City: 20016 - Washington, USA Other information about the project.