I am just learning R. I have developed a regression model with six predictor variables. While developing it, I found the relationships are not very linear. So, maybe because of this the predictions of my model are not exact.
Here is Headers of my data set:
1.bouncerate(To be predicted)
2.avgServerResponseTime
3.avgServerConnectionTime
4.avgRedirectionTime
5.avgPageDownloadTime
6.avgDomainLookupTime
7.avgPageLoadTime
Sample datasets:
28.57142857,4.132,0.234,0,0.505,0,14.168
42.85714286,3.356777778,0.090777778,0.077333333,0.459,0.105444444,14.78644444
0,3.372,0.1105,0.0015,0.425,0.1305,34.3425
33.33333333,3.583,0.218,0,0.385,0.649,11.816
66.66666667,2.438,0.234,0,0.3405,0,8.645
100,2.805,0.179666667,3.203666667,0.000333333,0.11,13.47066667
66.66666667,0.977,0,0.003,0,0,12.847
0,2.776,0,7.888,0,0,14.393
100,2.59,0.261,0,0.517,0,6.216
Here is the summary of my model:
Call:
lm(formula = y ~ x_1 + x_2 + x_3 + x_4 + x_5 + x_6)
Residuals:
Min 1Q Median 3Q Max
-125.302 -26.210 0.702 26.261 111.511
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 48.62944 0.27999 173.684 < 2e-16 ***
x_1 -0.67831 0.08053 -8.423 < 2e-16 ***
x_2 0.07476 0.49578 0.151 0.880143
x_3 -0.22981 0.06489 -3.541 0.000399 ***
x_4 0.01845 0.09070 0.203 0.838814
x_5 3.76952 0.67006 5.626 1.87e-08 ***
x_6 0.07698 0.01565 4.919 8.75e-07 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 33.76 on 19710 degrees of freedom
Multiple R-squared: 0.006298, Adjusted R-squared: 0.005995
F-statistic: 20.82 on 6 and 19710 DF, p-value: < 2.2e-16
plot with all single variable are below:
I have certain questions about this model:
- Is there any way to improve the accuracy of this model?
- Which of the values is most useful: residual standard error, degrees of freedom, multiple R-squared, adjusted R-squared, F-statistics, or p-values for choosing best model?
- Is it appropriate to use polynomial transformations with these data?
- In case I do use polynomial terms in my model, which degree is most appropriate?