Research And System Realization Of Impairment Prediction Of IPTV Based On Hadoop

With the continuous improvement of technology and the continuous development of Internet technology,IPTV is gradually changing people's lifestyle.However,due to the failure of the IPTV set-top box itself or the problem of network transmission jams,the user occasionally suffers from problems such as caton and flower screen when using the IPTV set-top box,resulting in a decrease in user experience.In order to enhance the user experience and reduce the loss of subscribers,operators hope to establish an IPTV failure prediction model based on the existing IPTV set-top box data and subscriber data to predict the IPTV failure so that can test and maintenance of IPTV set-top boxes in advance to solve the problems the user is facing and improve user experience.This dissertation is based on this issue.On the one hand,according to the conditional independence assumption of the traditional Naive Bayesian classification algorithm and the correlation coefficient between the information attribute of the data attribute itself and the data attribute and the decision attribute,the dissertation compares the ratio of the information gain to the overall information gain and as an integrated weight,a naive Bayesian classification algorithm based on comprehensive weighting is proposed,which is used as a classifier in forecasting impacted users of IPTV users.Experimental results show that the proposed algorithm has better classification performance and more stable performance than the ordinary NB algorithm,WNB-G algorithm and WNB-CC algorithm.On the other hand,since the IPTV dataset is an imbalanced dataset,this dissertation improves the traditional SMOTE algorithm based on the characteristics of IPTV data,and proposes an improved SMOTE oversampling algorithm to achieve the processing of unbalanced dataset.The algorithm first clusters a few samples by K-Means method and deletes the nearest subclass of centroids of the cluster after clustering.Then,in each cluster,the number of neighbors of the sample is used to classify the clusters again and remove the noise samples.Then according to the ratio of the input random number to the number of sub-class samples in the cluster,SMOTE resampling is selected in different subclass samples.Finally,we combine the algorithm with the weighted Naive Bayesian classification algorithm to establish the IPTV user failure prediction model.The experimental results show that the model is better than other models.In the aspect of system implementation,this dissertation uses Hadoop platform to establish the IPTV failure prediction system and implements an easy-to-operate front-end visual interface.Background mainly includes data storage,distributed computing,communications,several functions,front-end visual interface to provide login,permissions management,interactive features,display capabilities,management functions.