Distributed Denial of Services (DDOS) attacks affect the availability of WEB services for an indeterminate period of time, flooding company servers with fraudulent requests and denying requests from legitimate users, generating economic losses due to unavailability of services. rendered. For this reason, the scope of this document is to develop a DDOS attack detection prototype from machine learning (SVM2), which captures network traffic, filters HTTP3 headers, normalizes data based on operational variables : False Positive Rate, False Negative Rate, Classification Rate, and sends the information to the SVM for the respective training and detection tests, integrated with the statistical software for data mining WEKA4, allowing to effectively identify these anomalous behaviors in the layer above the session (OSI5 Reference Model), in order to increase the availability time of services. The experiment will allow to evaluate, validate and compare the technique of the prototype based on a supervised SVM model, against a traditional model based on rules such as SNORT (Snort, 2008).