Prediction of cumulative egg production in japanese quails by using linear regression, linear piecewise regression and MARS algorithm

Abstract

The study aims to predict the cumulative egg production of Japanese quails’ by using linear regression, linear piecewise regression, and multivariate adaptive regression splines algorithms including age at sexual maturity, weight at sexual maturity, average weight of the first ten eggs, and partial-egg records (20, 30, 40, 60, 80, 100, and 150 d partial-egg records). All the raw data were acquired from a total of 128 female quails. To compare prediction methods, the fit criterions of 15 different models were examined, moreover the models were compared with the most common criterions.
All prediction methods showed similar results, when the 40, 60, and 80 d partial-egg records included as independent variables in the models. Although the linear regression and the MARS algorithms inferred satisfying performance with 100 and 150 d of partial-egg records, the linear piecewise regression models gave a worse prophesying performance than others did. In conclusion, as an early (indirect) selection criterion, partial-egg records from d 100 can be successfully included as independent variable into the linear regression and MARS models to predict cumulative egg production.