CH11.Classification.pdf

# CH11.Classification.pdf - Chapter 11 Discrimination and...

• Notes
• 33

This preview shows pages 1–6. Sign up to view the full content.

Chapter 11. Discrimination and Classification 1. Introduction Discrimination and classification are multivariate techniques concerned with sep- arating distinct sets of objects (or observations) and allocating new objects to previously defined groups. Discrimination is rather exploratory while classification procedures are less ex- ploratory. In theses days, people use the term classification more often than discrimination. Classification is also called supervised learning. Example Group: { good credit, bad credit } Input variables: { age, education level, income , . . . } . The objective is to make a rule - a function of input variables, which discriminate the two groups, based on the given data, and, classify a new customer, who has only input variables, based on the rules constructed. 1

This preview has intentionally blurred sections. Sign up to view the full version.

2. Classification with Two Population (A) Fisher’s idea Two populations with common variance : π 1 : ( µ 1 , Σ ) , π 2 : ( µ 2 , Σ ) Objective : Observe X either from π 1 or π 2 , and specify a classification rule ( R 1 , R 2 ) or R 2 where X is from π 1 ” if X R 1 X is from π 2 ” if X R 2 with R 1 R 2 = R p , R 1 R 2 = ϕ . Idea of “maximally separating projection” (p.590) : arg max a ( E ( a X | π 2 ) E ( a X | π 1 )) 2 Var( a X | π ) = Σ 1 ( µ 2 µ 1 ) Note on “the scaling invariance”. (p.589) Classification with linear discriminant function (p.591) : R LD 2 : | a X a µ 2 | < | a X a µ 1 | for a = Σ 1 ( µ 2 µ 1 ) , i.e., ( µ 2 µ 1 ) Σ 1 X > 1 2 ( µ 2 µ 1 ) Σ 1 ( µ 2 + µ 1 ) . 2
Unknown parameter case (p.591) : X 11 , · · · , X 1 n 1 : r.s. from ( µ 1 , Σ ) X 21 , · · · , X 2 n 2 : r.s. from ( µ 2 , Σ ), indep. of X 1 j ( j = 1 , · · · , n 1 ) X : new observation from either π 1 or π 2 R LD 2 : ˆ a X > 1 2 ˆ a µ 2 + ˆ µ 1 ) ˆ a = (ˆ µ 2 ˆ µ 1 ) ˆ Σ 1 , ˆ µ i = ¯ X i ( i = 1 , 2) ˆ Σ = S pooled - maximum separation through t -statistic (p.590) : arg max a | a ¯ X 2 a ¯ X 1 | p n 1 1 + n 1 2 p a S pooled a = S 1 ( ¯ X 2 ¯ X 1 ) Example 11.4 (p.591) 3

This preview has intentionally blurred sections. Sign up to view the full version.

(B) Bayes and Minimax classification Two populations with known pdf’s : π 1 : pdf f 1 , π 2 : pdf f 2 Assume f 1 and f 2 are of continuous type. Costs (“losses” : terminology in decision theory) (p.581) : Classification (“action”) π 1 π 2 True π 1 0 c (2 | 1) π 2 c (1 | 2) 0 Expected costs (“risks”) ( c (2 | 1) P ( X R 2 | π 1 ) , c (1 | 2) P ( X R 1 | π 2 )) with R 1 = R c 2 . - misclassification probabilities for 0-1 loss. Minimax criterion : - minimize the maximum risk : minimize max { c (2 | 1) P ( X R 2 | π 1 ) , c (1 | 2) P ( X R 1 | π 2 ) } with respect to R 2 = R c 1 Bayes criterion (p.581, p.583) : Given “prior” probabilities p 1 = P ( π 1 ) & p 2 = P ( π 2 ), minimize the average (w.r.t. prior) risk (Bayes risk) c (2 | 1) P ( X R 2 | π 1 ) p 1 + c (1 | 2) P ( X R 1 | π 2 ) p 2 with respect to R 2 = R c 1 - “ECM on p.581” “TPM on p.583 for 0-1 loss” 4
Bayes classification rule (p.581) R B 2 : ( p 2 f 2 ( x )) / ( p 1 f 1 ( x )) > c (2 | 1) /c (1 | 2) where the l.h.s. (r.h.s.) is called the posterior odds (cost, respectively) ratio (p.584).

This preview has intentionally blurred sections. Sign up to view the full version.

This is the end of the preview. Sign up to access the rest of the document.
• Spring '16
• Bayesian probability, Prior probability, Bayes estimator

{[ snackBarMessage ]}

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern