Using Incremental Learning To Predict Coronavirus Infection Among Humans

Since the end of 2019,the pandemic of novel coronavirus has seriously affected public health and social order,and the prediction method based on machine learning can distinguish the infectivity and pandemic risk of coronavirus.At present,six kinds of coronaviruses(hcov-229 e,hcov-oc43,SARS Co V,HCo V-NL63,hcov-hku1 and mers COV)that infect human beings have been found.The virus genome sequence will change significantly over time,resulting in the decline of existing machine learning performance and the phenomenon of learning forgetting.Therefore,it is necessary to establish a coronavirus interpersonal Infection Prediction Model Based on incremental learning to realize the continuous monitoring of the infection risk of virus variants.Incremental learning model can learn new knowledge from new samples and retain old knowledge at the same time,which can effectively resist the learning forgetting phenomenon in the process of continuous training of neural network.This thesis implements a robust coronavirus infection risk prediction model based on incremental learning strategy.The main research contents include: 1)coronavirus infectious line prediction model based on spike protein.For spike protein data,one class SVM is used to classify groups,identify new virus groups,and continuously divide and predict new tasks.At the same time,in the framework of BP neural network model,the number of hidden layer nodes is increased with batch tasks,and the joint strategy of parameter sharing and knowledge distillation is used for incremental transformation.The results show that when the number of hidden layer nodes increases to 6,the prediction model achieves the best performance,IAC achieves the maximum value of0.9035 and BT achieves the maximum value of-0.0399,which effectively inhibits the learning and forgetting trend of the network model.2)Genome based coronavirus pandemic risk prediction model.The hierarchical clustering method is used to divide the groups,identify the new groups related to the virus pandemic,and divide the new prediction tasks.At the same time,the genome sequence segmentation and one-dimensional convolution are used to extract the characteristics of the virus genome,construct the coronavirus pandemic risk prediction model,and further carry out incremental transformation on the basis of BP neural network model.The results show that when the hidden layer node increases to 2,the prediction model achieves the best performance,IAC achieves the maximum value of 0.8748 and BT achieves the maximum value of-0.0763,which effectively suppresses the learning and forgetting trend of neural network model.Facing the epidemic characteristics of coronavirus,this thesis constructs incremental learning prediction models based on viral protein and viral genome data respectively to warn the epidemic risk of coronavirus from different angles.The results show that the prediction performance of the two models(IAC: 0.9035,0.8748)is close to that of data joint training(IAC:0.9236,0.9322),which is significantly better than that of neural network without knowledge distillation(IAC: 0.7764,0.7570).When compared with other incremental methods,it is better than sample-based method ESRIL(IAC: 0.8662,0.8252)and model parameter-based method CCL(IAC: 0.8853,0.8572),which has important public health application value.