Research On IPTV Users' Complaint Prediction Algorithm Based On Machine Learning

Nowadays,triple play is vigorously advancing in China.As the most suitable focus of triple play,IPTV(Internet Protocol Television)has a great potential,so the research on IPTV has become a hot spot.However,the operator's traditional operation and maintenance methods are mainly based on the user's complaints to exclude equipment failure.The poor timeliness,as well as the need for a large number of operation and maintenance personnel,leads to great redundancy,thus making the traditional method unsuitable to meet current needs.In order to ensure that users watch IPTV experience,IPTV business urgently need a more reasonable and more effective user prediction algorithm as a substitute.At the same time,with the rapid improvement of various computer performance,machine learning and social areas are also integrated tightly.In this thesis,we mainly study some key problems involved in IPTV users' complaint prediction based on machine learning.The main research contents are as follows:(1)This thesis presents a Relief feature selection algorithm based on F-Score and mutual information.Advantages of Relief feature selection algorithm include simplicity,fast operation,and the good performance of its selected feature subset.But its ability to choose redundant features is weak.Fisher Score is added to the Relief algorithm to further improve the advantages of Relief algorithm.At the same time,in order to reduce the redundant features,this thesis combines mutual information with Relief.The Experiments on multiple data sets show that the classification accuracy of the original algorithm is improved compared to the Relief feature selection algorithm based on F-Score and mutual information.(2)This thesis presents an AdaBoost algorithm with Weight Restrict and F1.AdaBoost classification algorithm is simple,stable and unlikely to overfit.However,it is highly possible that AdaBoost algorithm can give a large weight to the anomaly in the classification process,resulting in algorithm imbalance and the classification error rate are not suitable for the nonequilibrium data set.To solve this problem we proposed an AdaBoost algorithm with Weight Restrict and F1.The experimental results show that the algorithm can effectively improve the classification accuracy.The improved algorithm can be used to improve the classification accuracy.(3)In this thesis,the Relief feature selection algorithm based on F-Score and mutual information and the AdaBoost algorithm with Weight Restrict and F1 are applied to IPTV users'At present,there are still few researches on the combination of machine learning and IPTV at home and abroad.In this thesis,the data of IPTV is analyzed and pretreated to combine with machine learning.The experiment is carried out in IPTV data.The results show that the improved algorithm has improved prediction accuracy,compared with the original algorithm.