Comparison of predictive performance of data mining algorithms in predicting body weight in Mengali rams of Pakistan

Abstract

-The present study aimed at comparing predictive performance of some data mining algorithms (CART, CHAID, Exhaustive CHAID, MARS, MLP, and RBF) in biometrical data of Mengali rams. To compare the predictive capability of the algorithms, the biometrical data regarding body (body length, withers height, and heart girth) and testicular (testicular length, scrotal length, and scrotal circumference) measurements of Mengali rams in predicting live body weight were evaluated by most goodness of fit criteria. In addition, age was considered as a continuous independent variable. In this context, MARS data mining algorithm was used for the first time to predict body weight in two forms, without (MARS_1) and with interaction (MARS_2) terms. The superiority order in the predictive accuracy of the algorithms was found as CART > CHAID ≈ Exhaustive CHAID > MARS_2 > MARS_1 > RBF > MLP. Moreover, all tested algorithms provided a strong predictive accuracy for estimating body weight. However, MARS is the only algorithm that generated a prediction equation for body weight. Therefore, it is hoped that the available results might present a valuable contribution in terms of predicting body weight and describing the relationship between the body weight and body and testicular measurements in revealing breed standards and the conservation of indigenous gene sources for Mengali sheep breeding. Therefore, it will be possible to perform more profitable and productive sheep production. Use of data mining algorithms is useful for revealing the relationship between body weight and testicular traits in describing breed standards of Mengali sheep.

Body weight prediction using different data mining algorithms in Thalli sheep: A comparative study

“…They estimated R 2 (0.90), r (0.94), SD ratio (0.31), and MAPE (6.48) for the CHAID algorithm. The estimates obtained by Celik et al [26] were also better than those in the present study. Compared with the present R 2 (%) estimated for the CHAID algorithm, Mohammad et al [14] estimated a higher R 2 (%) value of 72% for the CHAID algorithm in the BW prediction based on WH, chest girth, BL, and breed in indigenous Pakistani sheep.…”

“…Their estimates were better than the results obtained from the present study. In a similar study conducted by Celik et al [26], they compared the predictive performance of different data mining algorithms in predicting the BW of Mengali rams in Pakistan. They estimated R 2 (0.90), r (0.94), SD ratio (0.31), and MAPE (6.48) for the CHAID algorithm.…”

CART ve CHAID Algoritmalarının Karayaka Koyun Islahında Kullanımı

“…The best algorithm produces minimum SD Ratio (Standard Deviation Ratio), MAD (Mean Absolute Deviation), RMSE (Root Mean Square Error) and coefficient of determination (R 2 ) criteria for goodness of fit. Standard Deviation Ratio, MAD and RMSE can be written as [24,25] ;…”

Description of the Relationships Between Different Plant Characteristics in Soybean Using Multivariate Adaptive Regression Splines (Mars) Algorithm

“…Among those, MARS is a statistically valuable tool that can capture relationship between sets of dependent and independent variables. There are other studies on MARS algorithm in agricultural sciences (Celik et al, 2017;Eyduran et al, 2017;Aytekin et al, 2018;Aksoy et al, 2018). MARS is an algorithm that can produce a powerful prediction equation in the response variable.…”