In order to maintain the normal operation of services, operators often need to monitor key performance indicators (KPIs) of network applications, promptly detect potential faults and make repairs. These key performance indicators are often time-series data, so this task can be regarded as anomaly detection for time-series data. The time-series data of network applications usually have characteristics such as large volume, low anomaly rate, and diverse anomaly patterns, with less labeled data. Therefore, unsupervised methods should be used for anomaly detection. We used four unsupervised models for anomaly detection in time-series data of web applications. To improve the detection performance, we combined them with Spectral Residual (SR). We evaluate the models on 6 publicly available real-world datasets, and find that models with SR achieved better results.