Você está na página 1de 10

Interpretation of PCA Results:

Through the scores and loadings in interesting projections

Interpretation of scores and loadings

Score plot : (row vectors)


Summary of the relationship among
the observations (or samples)

DATA
MATRIX

C
su om
pe ple
rim m
po ent
sa ary
bl &
e
!

Loading plot : (column vectors)


Summary of the variable(properties)
- a means to interpret
the patterns seen in the score plot

Example: Interpretation of scores and loadings


with semiconductor data
A=

Size of this data matrix, A, is 446 from 44 samples with 6 variables.

Example: Interpretation of scores and loadings


with semiconductor data (Continued)
1.

Scaling: auto scaling is used because variables have different units.

1.9425 0.5401
1.5621 0.1188

X = auto-scaled matrix of A = #
#

1.4335 0.1694
1.4335 0.5754

0.5887
0.5887

0.9224 2.5417
0.2332 0.0565

#
#
#
#

0.5887 0.6054 0.0789 0.7754


0.7544
0.5887 1.5220 1.4833
1.3495
0.2139

2. Calculation of covariance matrix (S) of matrix X

0.3011 0.3335
1
0.3011
1
0.0801

0.3335
0.0801
1
S = covariance matrix of X=
0.4168 0.4742 0.0964
0.2604 0.2574 0.0249

0.6447 0.6494 0.2588


Symmetrical ->eigenvectors are orthogonal!

0.4168
0.4742

0.6447
0.6494
0.0964 0.0249 0.2588

1
0.9122 0.6115
1
0.3674
0.9122

0.6115 0.3674
1
0.2604
0.2574

Matlab command: S=cov(X)

44 rows

Example: Interpretation of scores and loadings


with semiconductor data (Continued)
3. Calculation of eigenvalues and eigenvectors of covariance matrix (S)

0.7058 0.0786 0.3936 0.3928


0.0127 0.4316
0.0803 0.4572 0.4518 0.6541 0.0501 0.3874

0.0002 0.0245 0.5264 0.5302 0.6422 0.1693


P = eigenvectors of matrix S=

0.7551 0.0363 0.0697 0.2024 0.3560 0.5058


0.6253 0.1088 0.0702 0.4069 0.5065 0.4124

0.1791 0.7687 0.1041 0.2800 0.2164 0.4909


Loadings

0
0
0
0
0
0.0420
0
0.2204
0
0
0
0

0
0
0.5865
0
0
0
= eigenvalues of matrix S=

0
0
0
0.8700
0
0

0
0
0
0
1.2247
0

0
0
0
0
0
3.0563

Matlab command: [P, ]=eig(S)

Example: Interpretation of scores and loadings


with semiconductor data (Continued)
4. Calculation of scores matrix, T

1.9425 0.5401
1.5621 0.1188

X = auto-scaled matrix of A = #
#

1.4335 0.1694
1.4335 0.5754

0.5887

0.9224 2.5417
0.5887 0.2139 0.2332 0.0565

#
#
#
#

0.5887 0.6054 0.0789 0.7754


0.5887 1.5220 1.4833
0.7544
1.3495

0.7058 0.0786 0.3936 0.3928


0.0127 0.4316
0.0803 0.4572 0.4518 0.6541 0.0501 0.3874

0.0002 0.0245 0.5264 0.5302 0.6422 0.1693


P = eigenvectors of matrix S=

0.7551 0.0363 0.0697 0.2024 0.3560 0.5058


0.6253 0.1088 0.0702 0.4069 0.5065 0.4124

0.1791 0.7687 0.1041 0.2800 0.2164 0.4909

Size of score matrix, T = [X][P]


= (446)*(66)
= 446

Matlab command: T=X*P

Example: Interpretation of scores and loadings


with semiconductor data (Continued)
PC6

PC5

PC4

PC3

PC2

PC1

0.0812 0.8026 0.9228 0.0484 0.7721 3.3827


0.0048 0.6085 0.8208 0.6350 0.8050 0.8993

T = Scores = #
#
#
#
#
#

0.2372
0.0723
1.2781
0.3728
0.1069
1.2485

0.0224 0.3937 0.9300 0.1245 0.9149 2.4383


0.7058 0.0786 0.3936 0.3928
0.0127 0.4316
0.0803 0.4572 0.4518 0.6541 0.0501 0.3874

0.0002 0.0245 0.5264 0.5302 0.6422


0.1693
P = Loadings=

0.7551
0.0363
0.0697

0.2024
0.3560

0.5058

0.6253 0.1088 0.0702 0.4069 0.5065 0.4124

0.1791 0.7687 0.1041 0.2800 0.2164 0.4909

Birds of a feather flock together !!

Example: Interpretation of scores and loadings


with semiconductor data (Continued)
Score Plot
1.2247
100 = 20.41%
(total eigenvalues = 6)
0.8
VE

0.6

0.4
Loadings on PC 2

(20.41%) =

Loading Plot

atomic no.

radii

lattice const.

0.2

melting point

-0.2

-0.4

-0.6
-0.5

(50.94%) =

EN

-0.4

-0.3

-0.2

-0.1
0
0.1
Loadings on PC 1

0.2

3.0563
100 = 50.94%
(total eigenvalues = 6)

0.3

0.4

Example: Interpretation of scores and loadings


with semiconductor data (Continued)
High values of VE

By comparing the score and loading plot,


We can identify the relationships between
Samples and variables
0.8

Score Plot

Loading Plot
VE

0.6

Low value of EN

Loadings on PC 2

0.4

High values of radii

atomic no.

radii

lattice const.

0.2

melting point

-0.2

-0.4

EN

High value of EN
-0.6
-0.5

On the score plot,


Sit together: similar behavior between samples
ex.) Nitrides

-0.4

-0.3

-0.2

-0.1
0
0.1
Loadings on PC 1

0.2

0.3

0.4

On the loading plot,


Sit together: similar behavior between variables
ex.) VE, atomic no. lattice const., and melting point

Você também pode gostar