The Application Of Decision Tree And SVR Algorithm In Predicting The Occurrence Degree Of Wheat Aphids

In recent years, big data is well known all over the world. Agriculture as basic industry combined with big data will make the development of modern agriculture go into new areas. Wheat is one of the most important food crops in China, and Shandong province is one of the main grain producing areas in China. Sitobion avenae and Rhopalosiphum padi are the wheat pests in our country even the worldwide. Wheat aphids are r-strategy type pest which could not only induce direct damages by feeding on phloem but also transmit virus of plant diseases as efficient vectors, thus affecting the photosynthesis and respiration of wheat, resulting in a serious decline in yield and quality of wheat. However, due to the shortcomings of the past methods and data, it is difficult to solve the problem of the prediction of the occurrence degree, especially in the short-term forecasting. This study based on the concept of data, using decision tree and support vector regression(SVR) this 2 kind of machine learning algorithm, ignoring the sample is independent identically distributed hypothesis, analyses the relationship between 2003-2013 aphids' occurrence degree and ladybirds, parasitoids, daily maximum pressure, sunshine time number 18 kinds of variables in central area of Shandong. Making up and optimizing the short-term monitoring and warning model of wheat aphids for serving “unified prevention and cure” of wheat pests as well as scientific prevention and control rule of wheat aphids.1 Results of decision tree method analysisThrough training on the tree pruning of decision tree, finally got 10 strong relevant variables. Among them, the highest information gain rate was the sunshine hours(0.3782), followed by the lady beetles. According to the information gain rate, we construct of decision tree, and the variables' place in the decision tree diagram branch was decided by information gain rate. By analyzing, we got the correct value and the error value in the target value. The model's confidence level is 91.49% and it can run stably. By the fitting figure of decision tree predictive results the aphids, we know the grade 4 and grade 5 are far from the real values.2 Results of SVR methodThe real value has little difference with the predicted value in the analysis of SVR and the SVR training model for prediction of aphid occurrence level is more accurate. The regression value is 0.9216. The training set's mean absolute error(MAE), average relative error(MRE) and the root mean square error(RMSE) are 0.27, 0.76 and 1.01; the test set's the mean absolute error(MAE), average relative error(MRE) and the root mean square error(RMSE) are 0.41, 0.95 and 1.90. The predictive value and the true value have a higher degree of anatomists, which can make up the defect of C5.0 algorithm for forecasting the level 4 and level 5.The experiment showed that the accuracy of C5.0 machine learning algorithm was lower than SVR slightly in predicting the occurrence degree of wheat aphids especially for the higher fourth and fifth degree of the fourth and the fifth. SVR and C5.0 decision tree C5.0 were mutual complementation in forecasting the occurrence degree of wheat aphids, and their combination. These two models could be stable in predicting the occurrence degree of wheat aphids. We suggested that the models might play an important role in the monitoring and forecasting of the pest insects in wheat aphids.