ABSTRACT
Linear Discriminant Analysis (LDA) is a statistical model used to distinguish between two populations basing on a score derived from sample data. All the physical measurements on benign and malignant tumors represent two populations. LDA makes use of the information contained in the covariance matrices between groups and within groups. Fisher’s discriminant score makes use of linear regression model (with several variables) in which the dependent variable Y is dichotomous. Basing on the discriminant scores, a new subject whose group is unknown can be classified into one of the groups, usually with some error.
This chapter discusses the methodology of LDA with the help of real data. Practical instructions to carry out the analysis with SPSS are demonstrated. It is also shown that with the help of ROC curve analysis, a composite biomarker can be arrived at using LDA and its effectiveness is discussed. Methods of comparing the power of individual markers with LDA score a composite classifier are also discussed. (161)
