| Pikeperch(Sander lucioperca)is one of the most popular economic fishes in recent years.At present,there are reasonable farming technologies in China,but most of the existing farming methods are extensive farming,which cannot meet the market demand.Therefore,how to use the current advanced information technology to establish the growth and development model of pikeperch,change the traditional extensive breeding model,and achieve fine management has become a focus of attention in the aquaculture industry.This paper uses machine learning methods to study the growth and development data of pikeperch to establish a model of pikeperch growth.The main research includes the following aspects:First of all,The pikeperch farming data collected in this article has 24 characteristics,17040 items.There are redundant,missing,qualitative variables in the original data,and the dimensions of each feature are quite different.Therefore,preprocessing operations such as missing value filling,qualitative feature processing,and non-dimensionalization are performed.Analyze each feature in the data,realize the preliminary screening of the feature,and select 13 features(1 dependent variable and 12 independent variables).And analyze the correlation between the independent variable and the dependent variable by calculating the correlation coefficient,and then judge the rationality of feature selection.Second,Correlation analysis of 12 independent variables found that there is a certain correlation between the independent variables.and the KMO test and Bartlett spherical test are performed on the preprocessed data.The test results show that the data meets the basic requirements of applying the dimensionality reduction method.Therefore,factor analysis and principal component analysis are used to reduce the dimensionality of the preprocessed data.Both methods have finally determined that 5 factors can cover most of the information in the data.Then the processed data is divided into years and random methods to divide the training set and the test set,combined with two dimensionality reduction methods,based on BP neural network,support vector machine and XGBoost three machine learning methods to train the established model,When modeling,design different parameter values for each machine learning algorithm,so as to filter out the optimal parameter values required by each machine learning algorithm to establish a model,and use the test set data to verify the established model.the three values of mean-squared-error,mean-absolute-error,and degree of fit were used as the criteria for model evaluation.Finally,compare the pros and cons of different data set partitioning methods,different dimensionality reduction methods,and models built using three machine learning methods.The results show that,in addition to the large deviation between the prediction results of the model established by XGBoost and the true value after breeding,The prediction effects of models built by other combinations are good.Among them,three indicators based on randomly dividing the data set,using BP neural network and principal component analysis method to predict the results show the best comprehensively.The three indicators are MSE=534.5,MAE=17.78,and R2=0.987. |
