Research On Dimension Reduction And Olfactory Evaluation Model Based On Multi Physicochemical Characteristics Of Odor

With the rapid development of video and audio technology corresponding to vision and hearing for many years,"thousand mile eye" and "smooth ear" have been widely known.As another important sense organ of human beings,olfaction,which carries the heavy task of "ten thousand mile fragrance",has increasingly become the object of people's attention.Considering the complexity of olfactory perception,people have not been able to get a good mapping relationship between the physical and chemical properties of odor molecules and olfactory perception.The purpose of this paper is to try to build a good olfactory perception evaluation model,and use machine learning method to build the mapping relationship between the physical and chemical characteristics of odor and olfactory perception evaluation,so as to realize the prediction of olfactory perception.Firstly,this paper introduces two experimental data collection and preprocessing methods,namely olfactory perception evaluation and molecular physicochemical characteristics,and then focuses on the different feature selection methods,and constructs the physicochemical feature data set for dimension reduction based on these methods.The feature selection methods include random forest,lasso and PCA.Random forest is suitable for feature selection of high-dimensional data.By ranking the importance of features,effective feature information can be selected by voting;Considering the multicollinearity of data,Lasso algorithm is used to reduce the dimension of data;In addition,an efficient feature extraction method PCA is selected to do dimension transformation.A total of 43 feature spaces were screened out to pave the way for the prediction of olfactory sensation.Then,the ridge regression algorithm,which is commonly used in the regression model,is introduced and applied to the olfactory perception score prediction model.Through the experiment,it is found that the ridge regression olfactory perception prediction and evaluation model based on random forest feature space is better than the ridge regression olfactory perception prediction and evaluation model based on the other two feature spaces.In addition,this paper introduces the elastic network regression algorithm,studies each feature space,analyzes the prediction evaluation model,and obtains that the feature space most suitable for the algorithm is selected by PCA.Finally,through the comparative analysis of different regression models,the conclusion is that the elastic network learner combined with PCA feature space can significantly improve the olfactory prediction results,and the performance of this model is the best;The combination of ridge regression and random forest is the most stable.Based on different feature spaces,combined with the classical ridge regression algorithm and elastic network algorithm,this paper can improve the prediction accuracy of part of olfactory evaluation,and optimize a better olfactory evaluation model,which makes a certain contribution to the automatic evaluation of olfactory prediction,and also provides a feasible scheme for the realization of bionic olfactory.