The self-organizing map for determination of main features related to biological yield and yield of wheat

Abstract

Among various methods of artificial neural networks (ANNs) and learning algorithms, self-organizing map (SOM) is one of the most popular models. The aim of this study is to classify features influencing the biological yield and yield of wheat using SOM algorithm. In SOM, according to qualitative data, the clustering tendency of yield and biological yield of wheat were investigated using 11142 data from 16 features. Data was collected from the literatures on the subject of wheat in Iran that was existed in http://sid.ir website. Results showed that when biological yield was as output, K with soil pH, irrigation regime with 1000-kernel weight and organic content (OC) with grain/spike were related to each other closely. Moreover, grain/spike and OC had closer relationship to biological yield. In contrast, negative relationship was observed between soil pH (r=-0.47) and HI (r=-0.61) with biological yield. When wheat grain yield was output of SOM model, K with soil pH, and P with OC was related to each other closely. Overall, grain/spike, P and OC were much closer related to crop yield than other parameters. Similar to biological yield, labels map showed that data classified in three classes for wheat yield and the top four rows of U-matrix were placed in class A. A clear separation was observed among class A with B and C. The characteristics of each group in the study area showed that group 2 with 0.784 (kg/m 2) had the highest yield than group 1 (0.241 kg/m 2) and group 3 (0.401 kg/m 2) so that in group 2, the amount of P (0.003 kg/m 2), OC (0.47%), pH (7.78), rainfall (492.45 mm), grain/spike (43.71) and spike/m 2 (668.21) and HI (37.53%) were higher than the other groups and related to yield directly. Our results showed that among the yield components, grain/spike was the most important features contributing to grain yield than spike/m 2 and 1000-kernel weight using SOM.

Using self-organizing maps for determination of soil fertility (case study: Shiraz plain)

“…The SOM method is a powerful technique for soil fertility class determination. In fact it is one of unsupervised learning methods, which means that no human intervention is needed during the learning and little needs to be known about the characteristics of the input data (Lee et al 2007;Mokarram et al 2014;Bijanzadeh & Mokarram 2016). The SOM offers a solution to apply a number of visualizations linked together (Buza et al 1991).…”