The topic of Complex Event Processing or CEP is of high importance due its applicability to many areas in Computer Science. From human activity recognition to anomaly detection in computer networks, the task of processing large amounts of events and being in a position to interpret them accurately, is very useful across many areas of study. This process of pattern recognition is done by rules, which usually are manually defined and applied to data streams. This can be time consuming and generally is very complex when the event variables are numerous. In this paper we present a new approach for the generation of these kinds of rules based on machine learning techniques. Moreover, these rules are not only generated automatically, but most importantly, they are generated online based on the inputs and feedback in the system. This means that the rules change over time because the system is "learning", making this proposal more efficient and automated. The approach was implemented using Support Vector Machines and the TESLA query model, which its implementation yielded promising results