Agricultural Informatics. Группа авторов

Figure 1.1 Layers and connection of a feed-forward back propagation ANN [2].

Machine learning algorithms are reliable for decisions and can be integrated with statistics for implementation in applied machine learning [5]. Machine Learning is an emerging subject to expedite the release of new genotypes. There exist several uses of Machine Learning in maize breeding. Few of them can be enlisted as loci mapping based on quantitative traits, heterotic group assignment and selections based on genomes. Authors [6] implemented ANN to predict crop production. The parameters used for the task were related to soil properties such as pH, nitrogen, phosphate, potassium, organic carbon, calcium, magnesium, and sulfur, manganese, copper, iron, depth and climate parameters such as temperature, rainfall, humidity. Cotton, Sugarcane, Jawar, Bajara, Soybean, Corn, Wheat, Rice and Groundnut were the crops taken for experiment.

       1.1.1.2 Information Fuzzy Network

      Research experiments analyzed parameters for predicting crop yield. The study used aggressive neural network for prediction. Remote sensing was one among the used parameters for the work [7, 8]. The experiments implemented the flexible Neuro-fuzzy Inference system (ANFIS). The inputs to the ANFIS were moisture content available in soil, biomass information of the ground and repository organ. Yield was the output node for result. Limited data was used for designing the network to predict values for future. This is the challenge in prediction. Rearrangement of data was done by eliminating one year and using the remaining data. The estimation deviation was calculated and compared to the yield of the year that is left out. The procedure was applied recursively to all the years and averaged efficiency for prediction was obtained.

Experiments used Refs. [10–12] as stated in Figure 1.2. Hellenic sugar Industry used FINKNN for sugar production forecast based on the population of assessment. FINKNN is studied as K-nearest neighbor classifier that performs over the metric lattice of traditional convex fuzzy set. Results proved that FINKNN showed improved results for efficiency in forecasting.

Figure 1.2 Fuzzy cluster membership function representation in various field [9].

       1.1.1.3 Decision Trees

      Decision trees are the supervised learning techniques used in machine learning. The Decision Tree model is comprised of nodes, branches, leaf, terminal value, payoff distribution, rollback and certain equivalent and method. Decision trees have three variants of nodes and two variants of branches. Square is used to represent one of the decision points. The decision node is a point of choice for the tree. The decision nodes extend decision branches. Each node toward the end of the tree is called terminal node. There is an associated value associated with terminal node commonly referred to as payoff, outcome value. The terminal value is the measure of sequences of decisions or the resultant of the scenario in the tree. The construction of Decision tree algorithm is a two-step process that includes growth of tree and pruning. In the growth step the large decision tree is created, reduced and overfitting is removed. The second step does the tree pruning to reach a decision. The classification tree used for decision making is the obtained pruned tree [13].

      Prediction is influenced by various factors in agricultural explorations [14]. Variables associated with agronomics, application of nitrogen and weed control were used for machine learning and decision tree for yield forecasting and development of yield mapping. Both decision trees and ANN were implemented and results showed that greater accuracies were obtained from ANN results.

Authors [15] modeled productivity of soybean using decision trees. Data for climate in Bhopal district was collected for since 1984 to 2003 considering the climatic factors of evaporation, minimum and maximum temperature, humidity, and rainfall. The factors were studied for the production of soybean crop. Interactive Dichotomizer3 (ID3) algorithm was implemented on data. ID technique is based on information and two assumptions. Relative humidity was found to be a major parameter that affected soybean crop yield. Few rules were generated that helped know the lowest and highest prediction of soybean. The only one limitation of the model was that, the amount of yield production cannot be predicted [4, 16], as shown in Figure 1.3.

      The vast climatic diversity of India impacts the agricultural production in several parts of the country. Convenient decisions can be made by the farmers and policy inventors if the production can be forecasted in advance.