User behavior recordings are a handy tool to aid companies in understanding better how users interact with their websites and applications. With the increasing number of visitors, it is a complex task for employees monitoring the recordings to do a comprehensive review. While manually classifying every session is very time-consuming, creating a machine learning clustering model that automatically puts each session in a group with similar ones saves a lot of time and increases precision. An added benefit of using machine learning for clustering is finding outliers among the data entries. This thesis collaborates with Smartlook, which produces a tool that records visitors on its customers' websites and provides these videos with metadata and various analytical data.

• Find the most efficient way to process and clean big amounts of data provided by Smartlook, create an effective way to query data, extract valuable features, and rate data properties.
• Provide a comparison and a review of the approaches applied in research and theses for user behavior clustering.
• Use the prepared data to create clusters of sessions using various unsupervised machine learning models.
• Analyze and compare different models created using the groups of recordings, examine their similarities, and inspect outliers: detect if they can be considered anomalies.
• Provide statistical and visual results of the models with their comparisons based on the distribution of sessions into clusters and outliers.