| dc.description.abstract | In this study, it is aimed to apply to estimate the plant height of grass pea (Lathyrus sativus L.) plant, data mining methods MARS (Multivariate Adaptive Regression Splines) and Bagging MARS (Bootstrap Aggregating Multivariate Adaptive Regression Splines) algorithms. Plant height, dry leaf, dry stalk, wet leaf and wet stalk were considered in different lines as plant characteristics. Plants consist of Leo, Ela, 504, Line-17, 481, 563, 528, Coloratus, Karadag, İptas, Elazig Pop. and Gurbuz lines. In the experiment carried out in 12 lines and 3 replication plots, 36 plants were examined. Belonging to the MARS algorithm that predicts plant height, r (correlation coefficient), R 2 (determination coefficient), Adjust R 2, Standard deviation ratio (SDratio), Root-mean-square error (RMSE), Relative root mean square error (RRMSE), Performance index (PI), Mean error (ME), Relative approximation error (RAE), Mean absolute percentage error (MAPE) and Mean absolute deviation (MAD) values, were found 0.867, 0.752, 0.701, 0.498, 2.459, 5.937, 3.180, 0, 0.058, 4.733 and 1.941, respectively. The same statistics of the Bagging MARS algorithm were obtained as 0.901, 0.811, 0.811, 0.435, 2.148, 5.212, 2.878, − 0.056, 0.050, 4.101 and 1.735, respectively. To compute the prediction value, the samples were split in two sets, a training one and a test one (70% and 30% of the total samples). An iterative process (1000 iterations) was run with each successive loop designing the matrices randomly. In MARS and Bagging MARS algorithms, no overfitting problem was observed under a set of independent variables consisting of plant height and plant morphological characteristics. It was concluded that both algorithms are a good statistical tool in revealing the farmers’ trends in crop production at the observed location and produce important clues in increasing plant height. However, in the light of the findings of this study, the results of the Bagging MARS algorithm can be evaluated primarily. | en_US |
