Given the increased growing of Internet of Things networks and their presence in critical aspects of human activities, the security of devices connected to these networks becomes critical. Machine Learning approaches are becoming prominent as enablers for security solutions in computer networks due to its capacity to process traffic information in order to detect abnormal patterns which might represent attacks targeting infrastructures. In this paper, we propose to leverage Convolutional and Recurrent Neural Networks, two artifacts that have been successfully used in contexts such as image processing for pattern recognition, for the development of a security solution to be used in the context of Internet of Things. Our results show that this approach, when evaluated with a state-of-the-art data set, achieves around 99% of accuracy in the binary classification of attacks (i.e. normal traffic vs attack traffic) and 96% for multiclass classification (recognition of different types of attacks) accuracy. These results outperform proposals available in literature, showing a promising landscape for developing security solutions for IoT infrastructures.