Research On Opportunistic Network Link Prediction Method Based On Light Gradient Boosting Machine

The opportunistic network is an intermittent network which nodes are not constantly connected and often disconnected.It is a distributed system that communicates through the opportunistic connection caused by the frequent movement of nodes.It has the non-connected network topology,large scale,high dynamic,node mobility has certain social and non-periodic,and the opportunity communication has intermittent and time series characteristics.In recent years,the academic circles have paid more attention to it.Link prediction is one of the hotspots and difficulties in the research of opportunity network by studying the dynamic changes of node attributes and the network topology between nodes and estimating the possibility of future links between nodes.This thesis analyzes the characteristics of the opportunity network topology changing over time,introducing the existing link prediction methods at home and abroad and the wide application of the light gradient boosting machine in solving time series problems.According to the characteristics of nodes moving in the opportunistic network,the historical information is considered to improve the neighbor-based resource allocation index and the distance-based local path index,combing these two improved index to obtain the similarity index O_LS of the opportunistic network.In the case of analyzing the time series,the sliding time window is constructed after dividing the time slice,and the corresponding opportunity network similarity index O_LS is calculated to obtain the input of the prediction model.The decision tree is selected as the base learner and voting method as the integration strategy.The training model is optimized by grid search and cross-validation to determine the iteration number of the model and the number of tree leaf nodes.The link of the opportunistic network is predicted using a light gradient boosting machine model.This thesis uses two real-world datasets,ITC and INFOCOM,conducting experiments based on the Sklearn machine learning library,using precision and AUC as evaluation metrics to verify the effectiveness and stability of the similarity index O_LS,and to evaluate the performance of the opportunity network prediction models.The experimental results on the two datasets show that compared with the CN,PA,and RA similarity index prediction methods,the similarity index O_LS constructed in this thesis has better effectiveness and stability,and the constructed light gradient boosting machine and the predictive model has better predictive performance.