The increasing use of web applications and the popularity of Software-As-A-Service has created room for major vulnerability issues in systems which up until recently were “running” in restricted networks: information (sensitive or not) is now available on the internet. As a consequence, using appropriate software security procedures is the only way to protect it. Security checks must be performed in many and different layers, like the network layer, the OS layer, and also the application layer. In light of this, the objective of this diploma thesis is the design and development of a system that detects possible security attacks using machine learning algorithms. The goal is the use of machine learning algorithms to detect “good” and “bad” behaviors at the application layer. The analysis will be dynamic (at runtime) and a decision mechanism will be developed.