ANN Based Prediction Model Of Urticaria Syndrome And Implementation Of SVM Classifier

ObjectiveFor the purpose of providing help for clinical differentiation and treatment of urticarial TCM syndromes,the data of 397 urticarial cases,which were accordance with the criteria,were analyzed and mined with the method and approach of data mining,so as to explore the application of BP artificial neural network and support vector machine?SVM?technology in the research of urticarial TCM syndromes,and construct an unsupervised model of urticarial TCM syndrome prediction and a supervised classifier of urticarial TCM syndromes.MethodsThe data of 613 urticarial cases with Chinese medicine and treatment which were collected from the Traditional Chinese Medicine Knowledge Service and Sharing System,publications of clinical experience edited by famous traditional Chinese Medicine doctors and the Chinese databases such as Weipu?VIP?,Wanfang,CNKI,and others,were screened and cleaned.Among those data,the medical cases that meet the requirement of urticarial related records were dealt with normalization and input into the Excel form.Using Matlab2016a software,in the form of Matlab 2016a built-in function newff so as to adopt the artificial neural network toolbox of software and the LIBSVM toolkit,the establishment of card prediction model and the construction of SVM classifier are realized.Results1.Results of frequency analysisThe clinical symptoms were mainly presented as rash,pruritus and rash red,among which they were 382 times,291 times and 237 times respectively,accounting for 96.2%,73.3%and 59.7%respectively of the literature data of 397 cases.The tongue appearance was shown more as red tongue,light tongue and light red tongue,which were respectively 167,141 and 62 times,respectively accounting for 42.1%,35.5%and 15.6%of the 397 cases of literature.The tongue coatings were more presented as thin tongue coatings,white tongue coatings and yellow tongue coatings,respectively appeared 207times,181 times and 121 times,and respectively accounting for 52.1%,45.6%and 30.5%of those 397 cases.The pulse situations were showed more as thread pulses,rapid pulses and wiry pulse,which respectively were 146,122 and93 times,respectively accounting for 36.8%,30.7%and 23.4%of the 397 cases.Among the 7 common urticarial syndromes collected in this study,they were mostly seen as wind-cold fettering the exterior,viscera damp-heat syndrome and wind heat syndrome.They appeared 156 times,138 times and 97 times respectively,accounting for 39.3%,34.8%and 24.4%of 397 cases respectively.2.The result of constructing the predicating model of TCM syndromes.A data set of 35 elements and 397 samples was selected out from 397 samples with results more than 25 of symptoms,tongue,tongue coating and pulse According to the dimension of data set,the input layer number was set to35,and this data set was used as the input data of BP artificial neural network modeling.Three syndromes mostly appeared among the seven syndrome types which respectively were wind cold bundle,wind-heat invading syndrome,and damp-heat viscera syndrome,as the main research objectives.And the data of syndrome type in each sample,which form three data sets composed of 1element and 397 samples were extracted.The three datasets serve as the output layers of the three single output layer networks.In addition,the three data sets of three types of evidence were merged into a data set composed of three elements and 397 samples as the output layer to predict the multi-output layer.Of the input data,367 are training data and 30 are test data.Using newff function and artificial neural network toolbox,three different single-layer output-layer networks and a multi-layer output-layer network were modeled and outputted.The prediction accuracy of three single output layer network prediction models was 83.33%,80%and 80%,respectively.The average prediction accuracy is 81.11%,and the prediction accuracy of the multiple output layer network prediction model was 66.67%.3.Construction result of BP artificial neural network classifierThe data sets of 17 elements and 397 samples were selected from 397samples with symptoms,tongue,fur and pulse whose results were more than50.According to the dimension of data set,the input layer number was set to 17,and this data set was used as input data of BP artificial neural network classifier.Then the pre-classified sample label label.xlsx was extract,and taken from label=[1₁...1_n,2₁...2_n,3₁...3_n,4₁...4_n].The vector of was converted into a matrix composed of 0 and 1,and each column of the matrix was extracted separately to form 4 vectors of 0 and 1,Class1 to class 4.Among them,0 meat No and 1 meant Yes,that is,whether or not it fell into this category.The four class vectors were stored as Excel workbook files and imported into Matlab software as the output layer.Newff function and artificial neural network toolbox were used to realize BP artificial neural network urticarial syndrome classifier with 357 cases of input data as training data and 40 cases as test data.After adjusting the parameters,the accuracy rates of classification were 72.50%,90%,87.50%and 82.50%,respectively,within the constructed modes of wind cold bundle,wind-heat invading syndrome,damp-heat viscera syndrome and other syndromes.The average accuracy rate was 83.13%.The accuracy rate was fair enough,and the operation speed was fast.4.Construction result of SVM classifierMatlab 2016a software and the LIBSVM3.1 toolkit were applied to make the establishment.In order to ensure the comparability of the construction of SVM classifier,we also used the data of symptom,tongue,fur and pulse whose frequency was greater than or equal to 50,and considered them as input data.The best penalty function parameters and kernel function parameters were obtained by K-CV method,and the multi-type SVM classifier is constructed by using code and LIVSVM toolkit.The accurate rate of urticarial TCM syndrome classification was 92.5%.ConclusionThe clinical urticarial symptoms were mainly presented as rash,pruritus and color red,red tongue,light tongue,light red tongue,thin tongue coatings,white tongue coatings and yellow tongue coatings,and the pulse patterns mainly were thin tongue coatings,white tongue coatings and yellow tongue coatings.The pulses were more as thread pulses,rapid pulses and wiry pulses.The syndrome type commonly was wind cold bundles.In the construction of BP artificial neural network prediction model of urticarial syndrome type,the prediction accuracy of multiple artificial neural networks in single output layer was higher than that of single artificial neural network in multiple output layers.The multi-type SVM classifier was better than the BP artificial neural network classifier in the classification of measles TCM syndromes,and it can achieve good results.In this study,BP artificial neural network prediction model and multi-type SVM classifier of urticarial TCM syndrome were constructed,which to some extent filled the blank in the field of urticarial TCM syndromes researches.It provided a certain objective basis and help for the urticarial clinical treatment of,and also provides a new way of thinking and method for the study of urticarial TCM syndromes.