| Support vector machines and convolutional neural networks are classic machine learning algorithms widely used in image recognition,natural language processing,speech recognition,and other fields.Although both have demonstrated excellent classification performance in many aspects,they still face many challenges,such as low classification accuracy,long training time,overfitting,and lack of confidence levels.Particularly,the issues of low accuracy and lack of confidence level measurement are more prominent.When facing high-risk and low-tolerance problems,the algorithm’s prediction results may be inaccurate,uncontrollable,and unreliable.However,there are many such problems in real life,and how to solve the low accuracy and confidence level measurement problems of support vector machines and convolutional neural networks has become a topic that machine learning researchers need to overcome.To address this gap,this article proposes a support vector machine multi-classification algorithm and a convolutional neural network error rate controllable classification algorithm based on conformal predictors.For the multi-classification problem of the support vector machine via conformal predictors,we propose two multi-classification algorithms in this paper: One-Vs-Rest Support Vector Machine Algorithm via Conformal Predictors(OVR SVM CP)and One-Vs-One Support Vector Machine Algorithm via Conformal Predictors(OVO SVM CP),respectively.Firstly,the multi-classification problem is transformed into a binary classification problem,and the nonconformity measure is defined by the decision function.Then,the numerical simulation experiments are conducted for the two algorithms,which are compared with the OVO SVM,OVO LSSVM and OVO TWSVM,HSVM algorithm.Finally,the two algorithms are applied to six real data sets to test their predictive effects.The results of simulation experiment and real data application show that the predictive effects of the proposed two algorithms are relatively good,and have higher prediction accuracy than three other support vector machine algorithm.For the problem of unpredictable risk in the prediction results of convolutional neural networks,we introduce inductive conformal predictors,which can simultaneously solve the problems of confidence level measurement and low online learning efficiency.For the problem that conformal predictors is susceptible to non-balanced factors,we introduce mondrian conformal predictors and apply it to convolutional neural networks.We propose convolutional neural network classification algorithm based on mondrian inductive conformal predictors and study on three convolutional neural network models: Goog Le Net,Res Net,and Mobile Net.The analysis results show that the algorithm has a lower time complexity than the initial convolutional neural network and higher model training efficiency.It measures the degree of consistency among test samples and various classes based on non-conformal measure functions,making the output results highly reliable.Experimental results show that the proposed algorithm can effectively control the prediction accuracy of the neural network to achieve risk control and provide a reliable set of predicted values with a credible confidence level.Figure [22] Table [8] Reference [56]... |
