Predicting Number of Connections on Video Streaming Server with Machine Learning Approach

Abstract

Predicting a number of connection on the streaming server would be a useful parameter to improve the performance of the server. It can be also proved helpful to understand server behavior. Prediction can impact in improving resources on the server for providing quality video streaming. Streaming of videos from a server is resource hungry process and depends on many features like memory, processor, type of video codec, the bandwidth available and different network parameters (delay, jitter, drop, packet size). In this paper, we will use supervised learning technique on a captured dataset. Dataset is created from four different hardware based streaming server. Our first approach is to formulate a process for capturing data. We have calibrated a lab-based experiment setup on fifty mobiles and four different hardware based streaming server. Graph-based analysis on captured data is done to understand the behavior of the video streaming server. Performed the feature engineering to understand relationship among different features. Prediction is implemented using regression, and decision tree (DT) method. In regression we apply Linear regression, Ridge and Least Absolute Shrinkage and Selection Operator (LASSO). Basic DT and Random Forest (RF). So with this research work, we will show that it is possible to predict the number of connection on the server with exploiting server resources as features. Finally, compare algorithms these machine learning (ML) algorithms. In which RF prove to be best for prediction of connection on streaming server.