| At present,taxi application has become popular in our lives.Carpooling has also become a new way of saving money and protecting the environment.There are two main types of carpooling in taxi software.One is real-time carpooling,and the other is the previous appointment carpooling through sharing trips initiated in advance.Unlike the previous appointment carpooling that was born on the carpooling website,real-time carpooling is a kind of accompaniment with smart phones and mobile phone taxi application.This dissertation focuses on the results of carpooling under real-time carpooling scenarios.At present,researches on real-time carpooling mainly include carpool route matching,carpooling methods,and their developmental effects.This dissertation presents a new improved kNN algorithm for the problem of real-time carpooling prediction.According to the unbalanced distribution of dataset samples within and between classes,this algorithm mainly improves the traditional kNN algorithm in two aspects: First,according to the overall distribution imbalance of the sample,a noise neighbor detection method based on sample density is proposed based on density reachability in density clustering,which makes the determination of neighbors independent of the uneven distribution of samples.Second,based on the sample-like distribution disequilibrium,the average density of intra-class samples corresponding to each suspected neighbor is proposed as the density reachable detection distance threshold,so that the detection accuracy of noise neighbors is not affected by the inter-class distribution differences of samples.Finally,experiments on UCI public datasets prove the superiority of this algorithm in the unbalanced classification prediction problem of the sample distribution.It is verified on the actual carpool result dataset that the prediction effect of the algorithm has been significantly improved.This dissertation predicts the results of real-time carpooling to help users rationally arrange travel modes.At the same time,it also provides important preference information for drivers to grab orders.In the future,it can be applied to mobile phone taxi software platforms to provide users with more convenient and effective travel. |
