A Hybrid Approach for Web Traffic Prediction Using Deep Learning Algorithms

Abstract

The web traffic is a time series action and it is having its peak and low. Even though it can be predicted in different ways but as when we use the fastest growing time series databases, the prediction mostly will be accurate. Since time series is an important topic nowadays and it is having many applications, here we use those data to explore a study in the direction of web traffic prediction. In this paper, we propose a novel time series regression based hybrid model for web traffic prediction based on deep learning algorithms. This hybrid model is a combination of two well-known networks, Long Short Term Memory (LSTM) and Radial Basis Functional Networks (RBFNs). This is generated using ensemble stacking algorithm and is trained to make a final prediction using all the predictions of the LSTM and RBFNs as additional inputs. We choose the time series data on different Wiki pages and used the significant conventional evaluation metrics such as mean absolute error, mean squared error and mean absolute percentage error. The experiment section shows the measures and its comparisons; our prediction model provides less error by considering this random nature (change) for a large scale of data.