Wavelet-Based Hybrid Ensemble Forecasting for Accurate Web Traffic Prediction

This paper presents a novel approach for accurate web traffic prediction on the websites using a Wavelet-Based Hybrid Ensemble Forecasting method. The proposed method decomposes the web traffic time series into time-shifted discrete wavelets, enabling individual forecasting on each decomposed series. Four models - SARIMA, ARIMA, LSTM, and KNN - were evaluated, and their forecasts were combined in a hybrid ensemble model. Evaluation metrics such as accuracy percentage, MAE, RMSE, and MAPE were employed. The hybrid model exhibited superior performance, achieving an accuracy of 99.39%, along with the lowest MAE and RMSE values. Combining forecasts from multiple models significantly improved accuracy and reduced error rates in web traffic prediction.