Contemporary societies have become completely dependent on industrial control systems, which are responsible for many critical infrastructures taken for granted like water and electricity. Since these systems nowadays are utilizing distribution and interoperability involving routing data over public Internet, the need for countering cyber threats are emerging. This thesis pro-poses a novel approach for anomaly detection suitable for supervisory control and data acquisition (SCADA) systems. The anomaly detection model utilizes data recorded in a industrial control system and applies machine learning algorithms (random forest, multidimensional scaling plot) in order to capture a baseline profile of the data. An important aspect of the model is that it, by extracting higher level features, is capable of finding anomalies in traffic volume patterns within the network which complements the packet-level anomaly detection. Leveraging on the fundamental strengths of machine learning, the proposed model could be an important tool for enhancing security in these critical systems.