Ngày 4: Hồi qui tuyến tính

Phân tích tương quan

Việc 1. Đọc dữ liệu vào R

df = read.csv("C:\\Thach\\VN trips\\2025_2May\\Phuong Nam Institute\\Datasets\\Bone data.csv")

Việc 2. Đánh giá mối liên quan giữa cân nặng và mật độ xương cổ xương đùi

2.1 Vẽ biểu đồ đánh giá mối liên quan giữa cân nặng và mật độ xương cổ xương đùi

library(lessR)
## Warning: package 'lessR' was built under R version 4.3.3
## 
## lessR 4.3.9                         feedback: gerbing@pdx.edu 
## --------------------------------------------------------------
## > d <- Read("")   Read text, Excel, SPSS, SAS, or R data file
##   d is default data frame, data= in analysis routines optional
## 
## Many examples of reading, writing, and manipulating data, 
## graphics, testing means and proportions, regression, factor analysis,
## customization, and descriptive statistics from pivot tables
##   Enter: browseVignettes("lessR")
## 
## View lessR updates, now including time series forecasting
##   Enter: news(package="lessR")
## 
## Interactive data analysis
##   Enter: interact()
Plot(weight, fnbmd, fit = "loess", data = df)

## 
## >>> Suggestions  or  enter: style(suggest=FALSE)
## Plot(weight, fnbmd, enhance=TRUE)  # many options
## Plot(weight, fnbmd, fill="skyblue")  # interior fill color of points
## Plot(weight, fnbmd, MD_cut=6)  # Mahalanobis distance from center > 6 is an outlier 
## 
## Fit: Mean Squared Error, MSE = 0.016
## 

[ChatGPT]

PROMPT 1: “Bạn dùng gói lệnh ‘lessR’ để vẽ biểu đồ đánh giá mối liên quan giữa cân nặng (weight) và mật độ xương (fnbmd). Biểu đồ có đường biểu diễn mối liên quan thật sự giữa 2 biến số này. Bạn viết 1 câu để nhận xét gì về mối liên quan này”

ChatGPT:

scatter(weight ~ fnbmd, data = df, fit = TRUE)

PROMPT 2: Error in scatter(weight ~ fnbmd, data = df, fit = TRUE) : could not find function “scatter”

ChatGPT: “Xin lỗi về sự cố. Để vẽ biểu đồ và thêm đường biểu diễn mối liên quan thật sự giữa weight và fnbmd trong gói lessR, bạn có thể sử dụng hàm scatterplot thay vì scatter.”

PROMPT 3: “Trong gói lệnh ‘lessR’ không có lệnh ‘scatter’ hoặc ‘scatterplot’ (https://cran.r-project.org/web/packages/lessR/lessR.pdf). Biểu đồ tán xạ giữa 2 biến liên tục được vẽ bằng lệnh ‘Plot’”

ChatGPT: Cảm ơn bạn đã chỉ ra sự nhầm lẫn. Đúng là trong gói lệnh lessR, biểu đồ tán xạ giữa hai biến liên tục được vẽ bằng hàm Plot. Dưới đây là mã R chính xác để vẽ biểu đồ tán xạ giữa weight và fnbmd cùng với đường hồi quy:

library(lessR)

Plot(fnbmd ~ weight, data = df, fit = ‘lm’)

Câu nhận xét: “Mối liên quan giữa cân nặng và mật độ xương cho thấy một xu hướng tích cực, tức là khi cân nặng tăng, mật độ xương có xu hướng tăng theo, điều này có thể gợi ý rằng trọng lượng cơ thể có ảnh hưởng đến mật độ xương.”

2.2 Phân tích đánh giá tương quan giữa cân nặng và mật độ cổ xương đùi

Correlation(weight, fnbmd, data = df)
## Correlation Analysis for Variables weight and fnbmd 
##   
## 
## >>> Pearson's product-moment correlation 
##  
## Number of paired values with neither missing, n = 2121 
## Number of cases (rows of data) deleted: 41 
## 
## Sample Covariance: s = 1.269 
##  
## Sample Correlation: r = 0.581 
## 
## Hypothesis Test of 0 Correlation:  t = 32.882,  df = 2119,  p-value = 0.000 
## 95% Confidence Interval for Correlation:  0.552 to 0.609

[ChatGPT]

PROMPT 1: “Thực hiện phân tích đánh giá tương quan giữa cân nặng và mật độ cổ xương đùi”

ChatGPT:

Tính hệ số tương quan Pearson giữa cân nặng và mật độ cổ xương đùi

cor_result <- cor.test(df\(weight, df\)fnbmd, method = “pearson”)

In kết quả của phân tích tương quan

print(cor_result)

PROMPT 2: “Bạn thực hiện phân tích tương quan bằng gói lệnh ‘lessR’ được không?”

ChatGPT:

cor_result <- correlation(df\(weight, df\)fnbmd)

PROMPT 3: “Trong gói lệnh ‘lessR’ phân tích tương quan được thực hiện bằng lệnh Correlation (C viết hoa)”

ChatGPT: “Cảm ơn bạn đã chỉ ra điều này! Đúng là trong gói lệnh lessR, hàm để thực hiện phân tích tương quan là Correlation (với C viết hoa). Dưới đây là mã R sử dụng lessR để tính hệ số tương quan giữa weight và fnbmd:”

cor_result <- Correlation(df\(weight, df\)fnbmd)

Hồi qui tuyến tính

Việc 3. Đọc dữ liệu vào R

ob = read.csv("C:\\Thach\\VN trips\\2025_2May\\Phuong Nam Institute\\Datasets\\Obesity data.csv")

Việc 4. So sánh tỉ trọng mỡ giữa nam và nữ

4.1 Đánh giá phân bố của tỉ trọng mỡ

library(lessR)
Histogram(pcfat, fill = "blue", xlab = "Percentage of fat (%)", ylab = "Frequency",  data = ob)

## >>> Suggestions 
## bin_width: set the width of each bin 
## bin_start: set the start of the first bin 
## bin_end: set the end of the last bin 
## Histogram(pcfat, density=TRUE)  # smoothed curve + histogram 
## Plot(pcfat)  # Violin/Box/Scatterplot (VBS) plot 
## 
## --- pcfat --- 
##  
##        n   miss            mean              sd             min             mdn             max 
##      1217      0       31.604786        7.182862        9.200000       32.400000       48.400000 
## 
##   
## --- Outliers ---     from the box plot: 10 
##  
## Small       Large 
## -----       ----- 
##   9.2            
##   9.7            
##   9.8            
##  10.3            
##  10.3            
##  10.7            
##  11.0            
##  11.4            
##  11.7            
##  11.9            
## 
## 
## Bin Width: 5 
## Number of Bins: 9 
##  
##      Bin  Midpnt  Count    Prop  Cumul.c  Cumul.p 
## ------------------------------------------------- 
##   5 > 10     7.5      3    0.00        3     0.00 
##  10 > 15    12.5     26    0.02       29     0.02 
##  15 > 20    17.5     61    0.05       90     0.07 
##  20 > 25    22.5    128    0.11      218     0.18 
##  25 > 30    27.5    244    0.20      462     0.38 
##  30 > 35    32.5    338    0.28      800     0.66 
##  35 > 40    37.5    294    0.24     1094     0.90 
##  40 > 45    42.5    107    0.09     1201     0.99 
##  45 > 50    47.5     16    0.01     1217     1.00

4.2 Sử dụng kiểm định t

ttest(pcfat ~ gender, data = ob)
## 
## Compare pcfat across gender with levels F and M 
## Grouping Variable:  gender
## Response Variable:  pcfat
## 
## 
## ------ Describe ------
## 
## pcfat for gender F:  n.miss = 0,  n = 862,  mean = 34.672,  sd = 5.187
## pcfat for gender M:  n.miss = 0,  n = 355,  mean = 24.156,  sd = 5.764
## 
## Mean Difference of pcfat:  10.516
## 
## Weighted Average Standard Deviation:   5.362 
## 
## 
## ------ Assumptions ------
## 
## Note: These hypothesis tests can perform poorly, and the 
##       t-test is typically robust to violations of assumptions. 
##       Use as heuristic guides instead of interpreting literally. 
## 
## Null hypothesis, for each group, is a normal distribution of pcfat.
## Group F: Sample mean assumed normal because n > 30, so no test needed.
## Group M: Sample mean assumed normal because n > 30, so no test needed.
## 
## Null hypothesis is equal variances of pcfat, homogeneous.
## Variance Ratio test:  F = 33.223/26.909 = 1.235,  df = 354;861,  p-value = 0.016
## Levene's test, Brown-Forsythe:  t = -2.232,  df = 1215,  p-value = 0.026
## 
## 
## ------ Infer ------
## 
## --- Assume equal population variances of pcfat for each gender 
## 
## t-cutoff for 95% range of variation: tcut =  1.962 
## Standard Error of Mean Difference: SE =  0.338 
## 
## Hypothesis Test of 0 Mean Diff:  t-value = 31.101,  df = 1215,  p-value = 0.000
## 
## Margin of Error for 95% Confidence Level:  0.663
## 95% Confidence Interval for Mean Difference:  9.853 to 11.180
## 
## 
## --- Do not assume equal population variances of pcfat for each gender 
## 
## t-cutoff: tcut =  1.964 
## Standard Error of Mean Difference: SE =  0.353 
## 
## Hypothesis Test of 0 Mean Diff:  t = 29.768,  df = 602.015, p-value = 0.000
## 
## Margin of Error for 95% Confidence Level:  0.694
## 95% Confidence Interval for Mean Difference:  9.823 to 11.210
## 
## 
## ------ Effect Size ------
## 
## --- Assume equal population variances of pcfat for each gender 
## 
## Standardized Mean Difference of pcfat, Cohen's d:  1.961
## 
## 
## ------ Practical Importance ------
## 
## Minimum Mean Difference of practical importance: mmd
## Minimum Standardized Mean Difference of practical importance: msmd
## Neither value specified, so no analysis
## 
## 
## ------ Graphics Smoothing Parameter ------
## 
## Density bandwidth for gender F: 1.475
## Density bandwidth for gender M: 1.867

4.3 Sử dụng mô hình hồi qui tuyến tính

m.1 = lm(pcfat ~ gender, data = ob)
summary(m.1)
## 
## Call:
## lm(formula = pcfat ~ gender, data = ob)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -20.0724  -3.2724   0.1484   3.6276  14.8439 
## 
## Coefficients:
##             Estimate Std. Error t value            Pr(>|t|)    
## (Intercept)  34.6724     0.1826   189.9 <0.0000000000000002 ***
## genderM     -10.5163     0.3381   -31.1 <0.0000000000000002 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 5.362 on 1215 degrees of freedom
## Multiple R-squared:  0.4432, Adjusted R-squared:  0.4428 
## F-statistic: 967.3 on 1 and 1215 DF,  p-value: < 0.00000000000000022

Dùng gói lessR

library(lessR)
m.2 = reg(pcfat ~ gender, data = ob)
## 
## >>>  gender is not numeric. Converted to indicator variables.

m.2
## >>> Suggestion
## # Create an R markdown file for interpretative output with  Rmd = "file_name"
## reg(pcfat ~ gender, data=ob, Rmd="eg")  
## 
## 
##   BACKGROUND 
## 
## Data Frame:  ob 
##  
## Response Variable: pcfat 
## Predictor Variable: genderM 
##  
## Number of cases (rows) of data:  1217 
## Number of cases retained for analysis:  1217 
## 
## 
##   BASIC ANALYSIS 
## 
##               Estimate    Std Err  t-value  p-value    Lower 95%    Upper 95% 
## (Intercept)  34.672413   0.182622  189.859    0.000    34.314123    35.030703 
##     genderM -10.516344   0.338131  -31.101    0.000   -11.179729    -9.852959 
## 
## Standard deviation of pcfat: 7.182862 
##  
## Standard deviation of residuals:  5.361759 for df=1215 
## 95% range of residuals:  21.038669 = 2 * (1.962 * 5.361759) 
##  
## R-squared: 0.443    Adjusted R-squared: 0.443    PRESS R-squared: 0.441 
## 
## Null hypothesis of all 0 population slope coefficients:
##   F-statistic: 967.297     df: 1 and 1215     p-value:  0.000 
## 
## -- Analysis of Variance 
##  
##                df        Sum Sq       Mean Sq     F-value   p-value 
## Model           1  27808.311497  27808.311497  967.297285     0.000 
## Residuals    1215  34929.384159     28.748464 
## pcfat        1216  62737.695656     51.593500 
## 
## 
##   K-FOLD CROSS-VALIDATION 
## 
## 
##   RELATIONS AMONG THE VARIABLES 
## 
##           pcfat genderM 
##     pcfat  1.00   -0.67 
##   genderM -0.67    1.00 
## 
## 
##   RESIDUALS AND INFLUENCE 
## 
## -- Data, Fitted, Residual, Studentized Residual, Dffits, Cook's Distance 
##    [sorted by Cook's Distance] 
##    [n_res_rows = 20, out of 1217 rows of data, or do n_res_rows="all"] 
## --------------------------------------------------------------------------- 
##         genderM     pcfat    fitted      resid    rstdnt    dffits    cooks 
##    210        1  9.200000 24.156069 -14.956069 -2.801192 -0.148882 0.011020 
##    509        1 39.000000 24.156069  14.843931  2.780055  0.147758 0.010860 
##    179        1 38.700000 24.156069  14.543931  2.723523  0.144754 0.010420 
##    518        1  9.700000 24.156069 -14.456069 -2.706970 -0.143874 0.010300 
##    200        1  9.800000 24.156069 -14.356069 -2.688132 -0.142873 0.010150 
##    563        1 38.300000 24.156069  14.143931  2.648179  0.140749 0.009860 
##    318        1 10.300000 24.156069 -13.856069 -2.593980 -0.137869 0.009460 
##    972        1 10.300000 24.156069 -13.856069 -2.593980 -0.137869 0.009460 
##    388        1 10.700000 24.156069 -13.456069 -2.518700 -0.133867 0.008920 
##    203        1 11.000000 24.156069 -13.156069 -2.462262 -0.130868 0.008530 
##   1137        0 14.600000 34.672413 -20.072413 -3.766065 -0.128347 0.008150 
##    893        0 14.700000 34.672413 -19.972413 -3.747085 -0.127700 0.008070 
##    688        1 11.400000 24.156069 -12.756069 -2.387042 -0.126870 0.008020 
##    403        1 11.700000 24.156069 -12.456069 -2.330649 -0.123873 0.007640 
##    858        1 11.900000 24.156069 -12.256069 -2.293064 -0.121875 0.007400 
##    158        1 36.300000 24.156069  12.143931  2.271993  0.120755 0.007270 
##   1106        1 36.300000 24.156069  12.143931  2.271993  0.120755 0.007270 
##    827        1 36.000000 24.156069  11.843931  2.215637  0.117760 0.006910 
##    756        1 12.400000 24.156069 -11.756069 -2.199135 -0.116883 0.006810 
##    196        1 12.500000 24.156069 -11.656069 -2.180355 -0.115885 0.006690 
## 
## 
##   PREDICTION ERROR 
## 
## -- Data, Predicted, Standard Error of Prediction, 95% Prediction Intervals 
##    [sorted by lower bound of prediction interval] 
##    [to see all intervals add n_pred_rows="all"] 
##  ---------------------------------------------- 
## 
##         genderM     pcfat      pred   s_pred    pi.lwr    pi.upr     width 
##      2        1 16.800000 24.156069 5.369306 13.621929 34.690209 21.068280 
##      5        1 14.800000 24.156069 5.369306 13.621929 34.690209 21.068280 
## ... 
##   1209        1 26.400000 24.156069 5.369306 13.621929 34.690209 21.068280 
##      1        0 37.300000 34.672413 5.364869 24.146979 45.197847 21.050869 
##      3        0 34.000000 34.672413 5.364869 24.146979 45.197847 21.050869 
## ... 
##   1215        0 34.400000 34.672413 5.364869 24.146979 45.197847 21.050869 
##   1216        0 41.300000 34.672413 5.364869 24.146979 45.197847 21.050869 
##   1217        0 33.200000 34.672413 5.364869 24.146979 45.197847 21.050869 
## 
## ---------------------------------- 
## Plot 1: Distribution of Residuals 
## Plot 2: Residuals vs Fitted Values 
## ----------------------------------

Kiểm tra giả định

Cách đơn giản

par(mfrow = c(2, 2))
plot(m.1)

Dùng gói ggfortify

library(ggfortify)
## Loading required package: ggplot2
autoplot(m.1)

Mô hình

pcfat = 34.7 - 10.5*SexM

4.4 [ChatGPT]

PROMPT 1: “Dữ liệu ‘Obesity data.csv’ được đọc vào R và đặt tên là ‘ob’. Viết lệnh R để sử dụng kiểm định t để so sánh tỉ trọng mỡ (pcfat) giữa nam và nữ (gender). Viết 1 câu diễn giải kết quả”

ChatGPT:

t_test_result <- t.test(pcfat ~ gender, data = ob)

ttest(pcfat ~ gender, data = ob)

PROMPT 2: “Xây dựng mô hình tuyến tính để so sánh tỉ trọng mỡ giữa nam và nữ”

ChatGPT:

model <- lm(pcfat ~ gender, data = ob))

PROMPT 3: “Kiểm tra giả định của mô hình hồi qui tuyến tính trên”

“Có thể sử dụng lệnh hậu kiểm (postestimation) để đánh giá giả định của mô hình hồi qui tuyến tính. Ví dụ như lệnh ‘plot(model)’ sẽ cung cấp 4 biểu đồ để kiểm tra giả định (hoặc dùng lệnh ‘autoplot(model)’ của gói ‘ggfortify’”

ChatGPT:

Đúng rồi, bạn có thể sử dụng các lệnh hậu kiểm trong R để đánh giá các giả định của mô hình hồi quy tuyến tính một cách nhanh chóng và trực quan. Hai cách bạn đề cập (plot(model) và autoplot(model)) là rất hữu ích để kiểm tra các giả định quan trọng của mô hình, bao gồm:

Độc lập sai số (Independence of errors).

Đồng nhất phương sai (Homoscedasticity).

Phân phối chuẩn của sai số (Normality of errors).

Liên kết tuyến tính (Linearity).

  1. Sử dụng lệnh plot(model)

Lệnh plot(model) trong R sẽ tạo ra bốn biểu đồ có ích để kiểm tra các giả định của mô hình hồi quy tuyến tính. Đây là cách sử dụng lệnh này:

plot(model)

Các biểu đồ này sẽ bao gồm:

Biểu đồ 1: Residuals vs Fitted (Kiểm tra homoscedasticity và tính độc lập của sai số).

Biểu đồ 2: Normal Q-Q (Kiểm tra phân phối chuẩn của sai số).

Biểu đồ 3: Scale-Location (Kiểm tra homoscedasticity).

Biểu đồ 4: Residuals vs Leverage (Kiểm tra ảnh hưởng của các điểm dữ liệu có độ ảnh hưởng lớn).

  1. Sử dụng lệnh autoplot(model) của gói ggfortify Lệnh autoplot() từ gói ggfortify cũng tạo ra các biểu đồ hậu kiểm để đánh giá mô hình hồi quy tuyến tính. Nếu chưa cài gói ggfortify, bạn có thể cài đặt và sử dụng nó như sau:

library(ggfortify)

autoplot(model)

Việc 5. Đánh giá mối liên quan giữa cân nặng và tỉ trọng mỡ

5.1 Vẽ biểu đồ tán xạ

Plot(weight, pcfat, fit = "lm", data = ob)

## 
## >>> Suggestions  or  enter: style(suggest=FALSE)
## Plot(weight, pcfat, enhance=TRUE)  # many options
## Plot(weight, pcfat, fill="skyblue")  # interior fill color of points
## Plot(weight, pcfat, out_cut=.10)  # label top 10% from center as outliers 
## 
## 
## >>> Pearson's product-moment correlation 
##  
## Number of paired values with neither missing, n = 1217 
## Sample Correlation of weight and pcfat: r = 0.057 
##   
## Hypothesis Test of 0 Correlation:  t = 1.975,  df = 1215,  p-value = 0.049 
## 95% Confidence Interval for Correlation:  0.000 to 0.112 
##   
## 
##  Line: b0 = 29.222947   b1 = 0.043193    Fit: MSE = 51.471   Rsq = 0.003
## 

5.2 Sử dụng mô hình hồi qui tuyến tính

m.3 = lm(pcfat ~ weight, data = ob)
summary(m.3)
## 
## Call:
## lm(formula = pcfat ~ weight, data = ob)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -22.3122  -4.5234   0.8902   5.2695  16.9742 
## 
## Coefficients:
##             Estimate Std. Error t value            Pr(>|t|)    
## (Intercept) 29.22295    1.22370  23.881 <0.0000000000000002 ***
## weight       0.04319    0.02188   1.975              0.0485 *  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 7.174 on 1215 degrees of freedom
## Multiple R-squared:  0.003199,   Adjusted R-squared:  0.002378 
## F-statistic: 3.899 on 1 and 1215 DF,  p-value: 0.04855
library(ggfortify)
autoplot(m.3)

5.3 [ChatGPT]

PROMPT 1: “Xây dựng mô hình tuyến tính đánh giá mối liên quan giữa cân nặng (weight) và tỉ trọng mỡ (pcfat). Viết 1 câu diễn giải kết quả”

ChatGPT:

model_weight_fat <- lm(pcfat ~ weight, data = ob)

summary(model_weight_fat)

Ví dụ về câu diễn giải kết quả: Giả sử bạn nhận được hệ số 𝛽1=0.05 với p-value nhỏ hơn 0.05. Bạn có thể diễn giải như sau:

“Kết quả mô hình hồi quy tuyến tính cho thấy có mối liên quan dương giữa cân nặng và tỷ trọng mỡ. Cụ thể, mỗi tăng 1 kg cân nặng, tỷ trọng mỡ (pcfat) tăng trung bình 0.05%. Mối liên quan này có ý nghĩa thống kê với p-value < 0.05.”

Lưu ý rằng kết quả thực tế sẽ phụ thuộc vào dữ liệu của bạn, vì vậy các giá trị hệ số và p-value có thể khác nhau.

PROMPT 2: “Sử dụng gói ‘ggfortify’ để kiểm tra giả định của mô hình”

ChatGPT:

autoplot(model_weight_fat)

Việc 6. Đánh giá mối liên quan độc lập giữa cân nặng và tỉ trọng mỡ

6.1 Xây dựng và kiểm tra giả định mô hình

m.4 = lm(pcfat ~ weight + age + gender + height, data = ob)
summary(m.4)
## 
## Call:
## lm(formula = pcfat ~ weight + age + gender + height, data = ob)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -18.208  -2.543   0.019   2.582  15.706 
## 
## Coefficients:
##               Estimate Std. Error t value             Pr(>|t|)    
## (Intercept)  48.368722   3.505431  13.798 < 0.0000000000000002 ***
## weight        0.439169   0.015594  28.163 < 0.0000000000000002 ***
## age           0.056166   0.007404   7.585   0.0000000000000658 ***
## genderM     -11.483254   0.344343 -33.348 < 0.0000000000000002 ***
## height       -0.257013   0.023768 -10.813 < 0.0000000000000002 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 3.974 on 1212 degrees of freedom
## Multiple R-squared:  0.695,  Adjusted R-squared:  0.694 
## F-statistic: 690.4 on 4 and 1212 DF,  p-value: < 0.00000000000000022
autoplot(m.4)

6.2 Viết phương trình

pcfat = 48.4 + 0.44weight + 0.06age - 11.5genderM - 0.26height

6.5 [ChatGPT]

PROMPT: “Qua y văn bạn xác định các yếu tố có thể gây nhiễu (confounder) mối liên quan giữa cân nặng (weight) và tỉ trong mỡ (pcfat) là giới tính (gender), tuổi (age), và chiều cao (height). Hãy xây dựng mô hình đa biến đánh giá mối liên quan độc lập giữa cân nặng và tỉ trọng mỡ sau khi hiệu chỉnh cho các yếu tố gây nhiễu trên”

ChatGPT:

model_multivariate <- lm(pcfat ~ weight + gender + age + height, data = ob)

summary(model_multivariate)

Việc 7. Xây dựng mô hình dự báo tỉ trọng mỡ

7.1 Xây dựng mô hình dự báo tối ưu bằng phương pháp BMA

library(BMA)
## Loading required package: survival
## Loading required package: leaps
## Loading required package: robustbase
## 
## Attaching package: 'robustbase'
## The following object is masked from 'package:survival':
## 
##     heart
## Loading required package: inline
## Loading required package: rrcov
## Scalable Robust Estimators with High Breakdown Point (version 1.7-4)
yvar = ob[, c("pcfat")]
xvar = ob[, c("gender", "age", "height", "weight", "bmi")]
m.bma = bicreg(xvar, yvar, strict = FALSE, OR = 20)
summary(m.bma)
## 
## Call:
## bicreg(x = xvar, y = yvar, strict = FALSE, OR = 20)
## 
## 
##   3  models were selected
##  Best  3  models (cumulative posterior probability =  1 ): 
## 
##            p!=0    EV        SD        model 1      model 2      model 3    
## Intercept  100.0    5.26146  4.582901      7.95773     -0.79279      8.13735
## genderM    100.0  -11.25139  0.429659    -11.44430    -11.42764    -10.80625
## age        100.0    0.05259  0.008048      0.05497      0.05473      0.04715
## height      31.4    0.01759  0.028494        .          0.05598        .    
## weight      39.2    0.03102  0.042611      0.07921        .            .    
## bmi        100.0    1.01265  0.111625      0.89419      1.08852      1.08936
##                                                                             
## nVar                                         4            4            3    
## r2                                         0.697        0.696        0.695  
## BIC                                    -1423.06312  -1422.62198  -1422.49027
## post prob                                  0.392        0.314        0.294
imageplot.bma(m.bma)

[ChatGPT]

PROMPT: “Xây dựng mô hình dự báo tỉ trọng mỡ (pcfat) bằng phương pháp Bayesian Modelling Average (gói lệnh ‘BMA’)”

ChatGPT:

model_bma <- bic.glm(pcfat ~ weight + gender + age + height + bmi, data = ob, glm.family = gaussian())

summary(model_bma)

7.2 Kiểm tra giả định của mô hình

m.bma = lm(pcfat ~ gender + age + weight + bmi, data = ob)
summary(m.bma)
## 
## Call:
## lm(formula = pcfat ~ gender + age + weight + bmi, data = ob)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -18.225  -2.557   0.033   2.608  15.646 
## 
## Coefficients:
##               Estimate Std. Error t value             Pr(>|t|)    
## (Intercept)   7.957732   0.852500   9.335 < 0.0000000000000002 ***
## genderM     -11.444303   0.342565 -33.408 < 0.0000000000000002 ***
## age           0.054966   0.007395   7.433    0.000000000000199 ***
## weight        0.079207   0.028620   2.768              0.00573 ** 
## bmi           0.894194   0.080297  11.136 < 0.0000000000000002 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 3.963 on 1212 degrees of freedom
## Multiple R-squared:  0.6966, Adjusted R-squared:  0.6956 
## F-statistic: 695.7 on 4 and 1212 DF,  p-value: < 0.00000000000000022
library(car)
## Loading required package: carData
## 
## Attaching package: 'car'
## The following objects are masked from 'package:lessR':
## 
##     bc, recode, sp
vif(m.bma)
##   gender      age   weight      bmi 
## 1.878778 1.263468 5.609651 4.663425
autoplot(m.bma)

7.3 Viết phương trình

pcfat = 8.0 - 11.4genderM + 0.05age + 0.08weight + 0.89bmi

7.4 Xác định tầm quan trọng

library(relaimpo)
## Loading required package: MASS
## Loading required package: boot
## 
## Attaching package: 'boot'
## The following object is masked from 'package:car':
## 
##     logit
## The following object is masked from 'package:robustbase':
## 
##     salinity
## The following object is masked from 'package:survival':
## 
##     aml
## Loading required package: survey
## Loading required package: grid
## Loading required package: Matrix
## 
## Attaching package: 'survey'
## The following object is masked from 'package:graphics':
## 
##     dotchart
## Loading required package: mitools
## This is the global version of package relaimpo.
## If you are a non-US user, a version with the interesting additional metric pmvd is available
## from Ulrike Groempings web site at prof.beuth-hochschule.de/groemping.
ob$sex = ifelse(ob$gender == "F", 1, 0)
m.bma2 = lm(pcfat ~ sex + age + weight + bmi, data = ob)
calc.relimp(m.bma2, type = "lmg", rela = TRUE, rank = TRUE)
## Response variable: pcfat 
## Total response variance: 51.5935 
## Analysis based on 1217 observations 
## 
## 4 Regressors: 
## sex age weight bmi 
## Proportion of variance explained by model: 69.66%
## Metrics are normalized to sum to 100% (rela=TRUE). 
## 
## Relative importance metrics: 
## 
##               lmg
## sex    0.59317775
## age    0.06893066
## weight 0.09175463
## bmi    0.24613695
## 
## Average coefficients for different model sizes: 
## 
##                 1X         2Xs         3Xs         4Xs
## sex    10.51634414 11.71834412 11.80453842 11.44430262
## age     0.12768705  0.10445197  0.05168496  0.05496623
## weight  0.04319324 -0.05539405 -0.06907993  0.07920690
## bmi     1.03619023  1.50631405  1.54278433  0.89419395

[ChatGPT]

PROMPT 1: “Viết lệnh R để đánh giá tầm quan trọng của các biến số trong mô hình dự báo trên”

PROMPT 2: “Ý tôi là tầm quan trọng tương đối (relative importance) của các biến số trong mô hình dự báo”

PROMPT 3: “Cách làm này không chính xác. Bạn cần sử dụng gói lệnh ‘relaimpo’ để đánh giá tầm quan trọng tương đối của các biến”

ChatGPT: “Cảm ơn bạn đã chỉ ra điều này. Bạn hoàn toàn đúng rằng gói relaimpo là một công cụ phổ biến để đánh giá tầm quan trọng tương đối của các biến trong mô hình hồi quy. Gói relaimpo sử dụng phương pháp relative importance để tính toán mức độ đóng góp của mỗi biến vào mô hình hồi quy.”

relaimpo_result <- calc.relimp(model_lm, type = “lmg”, rela = TRUE)

print(relaimpo_result)

Việc 8. Ghi lại tất cả các hàm/lệnh trên và chia sẻ lên tài khoản rpubs