Application Of Decision Tree Algorithm In Soil Fertility Division

The soil nutrient status is an important index of soil quality, which relates to the crop growth, influences the structure, layout and benefit of agriculture production.In the early soil fertility evaluation, most researchers choose nitrogen, phosphorus, potassium and organic matter as evaluation indexes, and artificially determine the order of magnitude and weight coefficients of fertility indexes, this method of evaluation is subjective, also unscientific. The decision tree is a common classification model in the area of data mining, because of its less calculation, intuitive results and no need of human specified parameters, current studies have tried to apply the single variable decision tree to the evaluation of soil fertility.At present most decision tree algorithms use single variable attribute as its inspection attribute. As the amount of data increased, the scale of single variable decision tree is becoming lager, readability is worse, and it is more and more difficult to reflect the relationship between the properties. The theory constructs new multivariable decision tree based on traditional single variable decision tree around the data preprocessing and feature selection. The new algorithm is validated by Beijing suburban soil nutrient data, and compares with classical algorithm C4.5.(1)Aiming at the defaults of the traditional evaluation methods, the study applies attributes reduction of rough set to cut the attributes which disconnects to the target attribute, this can not only select attributes scientifically, but also delete the repeated data.(2) Aiming at the defaults of traditional single variable decision tree, such as large scale, poor readability, complex classification rules, putting forward a new multivariable decision tree algorithm. Firstly Extracting principal component using PCA, every principal component contains several soil nutrient factor, then construct multivariable decision tree algorithm using C4.5.This method can not only simplify scale of the decision tree, but also make the rules easier, it simplify the choice of the division points of the multivariable decision tree, reduce the constructing error and increase classification accuracy at the same time.(3)Experiment analysis. Comparing multivariate decision tree of this paper with the C4.5algorithm from structure complexity, node number, the accuracy of classification as well as time capability using Beijing suburban soil nutrient data.953agricultural measurements are taken, including11condition attributes,1decision attribute, after reduction we get7attributes, and then extract2principal components to construct multivariate decision tree. Experimental analysis results show that the depth of multivariable decision is half of tree C4.5algorithm, accuracy increased by10%, because the calculation of the study is bigger, running time is slower three times than C4.5algorithm. The numbers show that the study is feasible; it provides the theory and the practice model for precision agriculture development in the future.