Improving the Interpretation and Understanding of Web Traffic Time Series Forecasting with Statistical and Neural Network Models: A Machine Learning Approach

Abstract

The time series forecasting is a sequential event used for predicting future events based on historical time-stamped data. It relies on various statistical models to fit into the historical time-stamped data and uses them to predict future observations. Web traffic is a major issue as this can cause setbacks to the workings of major websites. Mainly wikipedia gives a decent growth and it contributes to economical growth. Since the volume of users visiting a website has increased exponentially these days, the management of web traffic is an important consideration for the flawless functioning of a website. Statistical models such as ARIMA, ETS, and NNAR play a vital role in prediction of unknown data from the historical data. The proposed model is intended to reduce RMSE and there by maintaining accuracy as well. The proposed model is developed by extracting important features using hybrid feature extraction methods. Then those features are used to predict the time series by feeding optimized LSTM model and finally compared with statistical models.