There are many projects today where data is collected automatically to provide input for various data mining algorithms. A problem with freshly generated datasets is their unsupervised nature, leading to difficulty in fitting predictive algorithms without substantial manual effort. One of the first steps in dataset preparation and mining is anomaly detection, where clear anomalies and outliers as well as events or changes in the pattern of the data are identified as a precursor to subsequent steps in data mining. In the research presented here, we provide a multi-step anomaly detection process which utilises different combinations of algorithms for the most accurate identification of outliers and events.