The advent of the Internet of Things (IoT) augurs new cutting-edge applications in modern life such as smart cities and smart grids. These applications require protocols more efficient for ensuring the reliability of data communications in the IoT networks. Many works state that IoT cannot meet their demands without application protocols improvement with Artificial Intelligence (AI) as IoT are expected to generate unprecedented traffic giving IoT researchers access to data that can help in studying and analyzing the demands and develop application protocols conceptions to meet the requirement of IoT applications. In literature, several works introduced AI in some layers of the TCP/IP model including wireless communication and routing. In this article, an evaluation of application protocols HTTP, MQTT, DDS, XMPP, AMQP, and CoAP has been presented; and subsequently, the power consumption prediction of MQTT and COAP based on the linear regression model is analyzed, in order to enhance data communications in IoT applications.