First Model, Ta-Feng

模型訓練與測試流程

Fig-1: The First Model

---

Loading & Preparing Data

Sys.setlocale("LC_ALL","C")

[1] "C"

library(dplyr)
library(ggplot2)
library(caTools)

Loading Data

rm(list=ls(all=TRUE))
load("data/tf2.rdata")

Spliting for Classification

TR = subset(A, spl)
TS = subset(A, !spl)

---

• 將顧客資料分成訓練資料及測試資料
• 利用訓練資料來製作模型，並且預測測試資料看此模型準不準

Classification Model

rr pred = predict(glm1, TS, type=) cm = table(actual = TS$buy, predict = pred > 0.5); cm

       predict
actual  FALSE TRUE
  FALSE  3730  873
  TRUE   1700 2273

rr acc.ts = cm %>% {sum(diag(.))/sum(.)}; acc.ts # 0.69998

[1] 0.69998

rr colAUC(pred, TS$buy) # 0.7556

                 [,1]
FALSE vs. TRUE 0.7556

• 檢視此模型，我們可以查看各個X對於Y的顯著程度
• AIC 越小越好
• 檢視acc , AUC

---

Regression Model

rr lm1 = lm(amount ~ ., TR2[,c(2:6,8:10)]) summary(lm1)

Call:
lm(formula = amount ~ ., data = TR2[, c(2:6, 8:10)])

Residuals:
    Min      1Q  Median      3Q     Max 
-1.8330 -0.2281  0.0485  0.2810  1.6424 

Coefficients:
              Estimate Std. Error t value     Pr(>|t|)    
(Intercept)  1.1403704  0.0504979   22.58      < 2e-16 ***
r            0.0000702  0.0003090    0.23       0.8203    
s            0.0001173  0.0003123    0.38       0.7072    
f            0.0256836  0.0017965   14.30      < 2e-16 ***
m            0.5045943  0.0372711   13.54      < 2e-16 ***
rev          0.0450307  0.0360945    1.25       0.2122    
ageB         0.0737926  0.0251165    2.94       0.0033 ** 
ageC         0.1204660  0.0230651    5.22 0.0000001800 ***
ageD         0.1264592  0.0227496    5.56 0.0000000279 ***
ageE         0.1382214  0.0232522    5.94 0.0000000029 ***
ageF         0.1085828  0.0242698    4.47 0.0000077690 ***
ageG         0.0787808  0.0264917    2.97       0.0029 ** 
ageH         0.0703242  0.0312462    2.25       0.0244 *  
ageI         0.0694822  0.0321119    2.16       0.0305 *  
ageJ        -0.0284007  0.0282282   -1.01       0.3144    
ageK         0.1124434  0.0395589    2.84       0.0045 ** 
areaB        0.0789586  0.0435321    1.81       0.0697 .  
areaC        0.0375241  0.0353641    1.06       0.2887    
areaD       -0.0111101  0.0371762   -0.30       0.7651    
areaE        0.0111809  0.0325803    0.34       0.7315    
areaF        0.0147066  0.0328141    0.45       0.6540    
areaG        0.0249228  0.0349567    0.71       0.4759    
areaH        0.0105550  0.0388962    0.27       0.7861    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 0.422 on 9246 degrees of freedom
Multiple R-squared:  0.291, Adjusted R-squared:  0.289 
F-statistic:  172 on 22 and 9246 DF,  p-value: <2e-16

• 檢視此預測模型
• 斜率的+/-表示正/負相關，大小表示對應變數影響程度
• R2表示此模型能夠解釋的變異程度
• 星號代表顯著的自變數

rr r2.tr = summary(lm1)\(r.sq SST = sum((TS2\)amount - mean(TR2\(amount))^ 2) SSE = sum((predict(lm1, TS2) - TS2\)amount)^2) r2.ts = 1 - (SSE/SST) c(r2.tr, r2.ts)

[1] 0.29099 0.25760

• 即總變異(SST)=已解釋變異(SSR)+ 未解釋變異(SSE)

---