A thesis Submitted to
Council of the Faculty of Management and Economics at
the University of Karbala As part of the requirements for a
Master’s Degree in Statistics Science

By
Maryam M. A. Al-khazali
Supervised By
Asis. Prof. Dr. Shrook A.S.AL-Sabbah

Abstract
The discriminate analysis is a statistical method based on a sample of observations taken from known societies to build a base that can help in identifying the society to which the new observations belong based on variables with discriminatory characteristics and applied to diagnose three types of tumors, namely breast, vertebral column (bone) and the respiratory system (lung) tumors as their prevalent in our society is abundance nowadays.
For the purpose of studying this subject, the values of
observations were recorded for five variables (sex, age, occupation, prognosis and duration of hospitalization)
for 270 patients (random samples) and they were studied
in four statistical models. The models were (linear discrimination function) and nonlinear (quadratic
discrimination function and logistic discrimination
function) and non-parametric model (nearest neighbor
function).
The statistical results were derived from the
statistical program Stata as well as the statistical
package SPSS to classify the tumors to which group they
belong. The classification error criterion was used as a
criterion for comparison between the four models, to the
best model with the least error of invention in possible.
It was found that the nearest neighbor model was
superior to the rest of the models in terms of lack of
classification of the wrong as compared to other models.