Method for predicting web server load based on wavelet analysis of web logs and web traffic

Abstract

The article is devoted to the study and development of a method for web server load prediction using wavelet analysis as a tool for preprocessing weblog and web traffic data. The relevance of the research lies in the need to ensure stable performance of web services under the conditions of increasing intensity and uneven distribution of workload. The proposed approach is based on the application of the discrete wavelet transform (DWT), which provides a multilevel decomposition of time series and makes it possible to capture both global trends and local fluctuations in traffic. Such decomposition increases noise resistance and facilitates the detection of hidden patterns in the data. The preprocessed time series are further applied in deep learning models, particularly in recurrent neural networks such as LSTM or the Prophet framework, which demonstrate high capability in modeling nonlinear dependencies and long-term correlations. An experimental comparison was carried out to evaluate the effectiveness of load prediction with and without wavelet-based filtering. The results confirmed a reduction in the mean prediction error, improved request balancing within the cluster, and increased overall fault tolerance of the system. The obtained findings have practical value for the design of intelligent resource management systems and can be integrated into monitoring solutions to enhance the efficiency and reliability of web services operating under dynamic load conditions