Applications of Machine Learning Algorithms in Nitrogen Fertilizer Management of Triticale

In this study, a new classification technique is proposed to distinguish the appropriate one from four different nitrogen (N) fertilizer doses (0, 40, 80, and 160 kg ha⁻¹) using six triticale cultivars. In the classification phase, nine yield featuresfrom 30 plants of the same cultivar were measured, that is, each dose or class has 30 feature vectors consisting of ninefeatures. Next, six triticale cultivars were classified for each dose of N fertilizer separately by using 30 feature vectorsbelonging to each dose. Similarly, the same classification task was repeated by using all feature vectors taken from fourdoses of N fertilizer. What makes this study novel is the classification process of six triticale cultivars by taking into accounttheir characters based on different doses of N fertilizer. The classification tasks were conducted by applying CommonVector Approach, Support Vector Machine, k-Nearest Neighbor, and Decision Trees algorithms. While satisfactory resultswere obtained from the training sets for all cases, the test set accuracy is relatively lower for the classification of four dosesof N fertilizer and six cultivars since features extracted from different doses of N fertilizer for the same cultivar are close toeach other. Furthermore, the number of feature vectors is insufficient to classify classes efficiently. Interestingly, when thecommon information of the classifiers was extracted with the biplot technique, useful results were obtained in selectingappropriate N doses for several triticale varieties. Combined with the results of future comprehensive studies, applicableresults for the agricultural sector can be proposed.