In the era of rapidly increasing web traffic and growing cybersecurity threats, effective prediction of server load and timely detection of anomalies play a crucial role in ensuring the reliability and security of web infrastructures. Traditional forecasting methods, such as ARIMA and exponential smoothing, often fail to capture short-term spikes and anomalies in traffic behavior, especially during cyberattacks like DDoS. Neural networks, particularly Long Short-Term Memory (LSTM) models, demonstrate improved accuracy in time series forecasting but remain sensitive to noisy data. This paper proposes the application of Wavelet Neural Networks (WNN) for predicting anomalous traffic on web servers. Wavelet decomposition is employed to separate traffic into low- and high- frequency components, enabling the detection of both long-term trends and short-term fluctuations. The WNN model is trained on preprocessed server log data and evaluated using standard error metrics (MAE, RMSE, MAPE) for forecasting, as well as precision, recall, and F1-score for anomaly detection. Experimental results show that WNN outperforms traditional methods and standalone LSTM models in capturing short-term spikes and improving anomaly detection accuracy. The findings highlight the potential of integrating WNN into real-time monitoring and cybersecurity systems, enhancing the resilience of web servers against cyber threats and ensuring more efficient resource allocation..