Mobile broadband networks are increasingly becoming more popular, thus increasing the need for quality monitoring and troubleshooting such systems. In the case of a malfunctioning network, the measured metrics get affected, indicating that an unexpected event — an anomaly — has occurred. This thesis deals with detecting time periods in which anomalies occurred. It is a problem of supervised learning where each measurement instance is marked as either anomalous or normal. We used two methods to search anomalies. The Concept drift method searches for anomalies, while the convolutional neural network also attempts to detect the size of anomaly regions. The latter method used the principle of time series classification. From one time series, we build several shorter series, which are then used as input attributes for convolutional neural networks. We evaluate the methods’ ability to detect and correctly determine the time period affected by anomalies on a real-world MBB measurement trace. Experiments show that the most appropriate neural network model predicts anomaly zones with a 73 % F1 score.