Application Of Least Squares Support Vector Machine Regression In Scoring Wine Quality

This article uses machine learning methods to explore how to score wine quality more accurately.The LS-SVR model based on variable selection was used to score the wine quality.The model was improved based on the SVR model,which reduced the complexity of the operation.In this paper,two examples are used to verify the effectiveness of the constructed LS-SVR model.First of all,this paper explored the data characteristics of the data set and conducted corresponding analysis and processing.The variables were selected according to the characteristics of the physicochemical indicators of the wine.The methods of GBRT,stepwise regression and lasso regression were successively used.Then,different models are applied for comparative analysis.For the red wine data set,the LS-SVR model,the LS-SVR model based on genetic algorithm and the RBF neural network model based on genetic algorithm were established,and their estimated values and errors were output respectively.Finally,the error of LS-SVR between the total variable set and the feature set is compared to prove the effectiveness of feature set construction.Comparing the error of genetic algorithm before and after optimization,the necessity of parameter adjustment is proved.Comparing the error between RBF neural network regression model and the LS-SVR model,it is proved that the evaluation effect of LS-SVR is better.The research proves that the MAPE values of the model built in this paper are all within 10%,which indicates that each model has achieved good prediction results,but the LS-SVR model has the lowest error and the best evaluation result.For white wine data sets,the LS-SVR model performs better than the RBF neural network model.In addition,the genetic algorithm is used to explore sigma and gama parameters in the LS-SVR model in a certain range to find a better parameter combination.It is found that the LS-SVR model is a better method,which opens up a new way for wine quality evaluation and can save the cost of studying wine quality to some extent.