ABSTRACT
In order to solve the problem of missing data from anthropometric database, this paper proposes a weighted regression method that can better repair the database. Through data classification and correlation analysis, least squares method and weighted regression method optimal setting and fitting effect analysis, the proposed “algorithm” was used to predict three data samples with different missing degrees. The results show that (1) the optimal setting for the least squares method is when the X and Y counts are both 1, and for the weighted regression method,is when the X and Y counts are both 1 and the weight decay factor is 90%. (2) The individual comparison effect indicator T ≤ 1 and the overall comparison effect indicator G = 0.68 < 1 both indicate that the weighted regression method has a better fitting performance. (3) The mean errors of the least squares method for the three samples with small missing data were 3.6, 1.16, and 1.11 times that of the weighted regression method. For with medium missing data, were 1.27, 1.16, and 1.05 times. For with large missing data,were 1.15, 1.42, and 1.25 times. In summary, the weighted regression method can provide better prediction results.
CO EDITOR-IN-CHIEF:
ASSOCIATE EDITOR:
Nomenclature
E | = | Error mean |
G | = | The overall comparison effect indicator |
Q | = | The mean square error |
R2 | = | The difference between the predicted and original data indicator formula |
RP2 | = | The best-fit effect indicator of the least square method |
RT2 | = | The best-fit effect indicator of the weighted regression method |
S | = | Sample number |
T | = | The comparison effect indicator |
= | The Pearson correlation coefficient | |
τ | = | The weight decay rate |
= | The mean square | |
wi | = | The weight decay factor |
Disclosure statement
No potential conflict of interest was reported by the author(s).