The future of the Internet of Things (IoT) is envisaged to consist of a high amount of wireless resource-constrained devices connected to the Internet. Moreover, a lot of novel real-world services offered by IoT devices are realised by wireless sensor networks (WSNs). Integrating WSNs to the Internet has therefore brought forward the requirements of an end-to-end quality of service (QoS) guarantees. In this thesis, a QoS framework for integrating WSNs with heterogeneous data traffic is proposed. The concept of Adaptive Service Differentiation for Heterogeneous Data in WSN (ADHERE) is proposed based on the varying QoS factors and requirements analysis of mixed traffic within an IoT-based WSN. The objective of the QoS framework is to meet the requirements of heterogeneous data traffic in the WSN - in the domain of timeliness and reliability. Another objective is to implement an adaptive QoS scheme that can react to dynamic network changes. This thesis provides the literature analysis and background study for integrating a WSN which contains heterogeneous data traffic with the Internet. In the discussion of network modelling and implementation tools for the testing, this thesis provides an insight into the different tools that are available and their ability to investigate the concept of service differentiation among heterogeneous traffic within the IoT-based WSN network. Furthermore, the major components of ADHERE are presented in the Concept chapter. The major components are: a heterogeneous traffic class queuing model that encompasses a service differentiation policy, a congestion control unit and a rate adjustment unit that supports the adaptive mechanism. Network modelling and the simulation of an ADHERE QoS framework which is carried out primarily using the network simulation tool, Riverbed Modeler, are also presented. Additionally, a proposed co-simulation between Riverbed Modeler and MATLAB is introduced, which aims to provide a seamless QoS monitoring using the ADHERE concept. The simulation results suggest that real-time traffic achieves low bound delay while delay-tolerant traffic experiences a lower packet drop. This indicates that the needs for real-time and delay-tolerant traffic can be better met by treating both packet types differently using ADHERE. Furthermore, a verification and added-value to the ADHERE QoS model using a neural network is also presented. The learning capabilities in ADHERE optimise the QoS framework’s performance by accommodating the QoS requirements of the network through the unpredictable traffic dynamics and when complex network behaviour takes place. Before concluding the thesis, the implementation of ADHERE QoS as a use-case on a physical test environment is also discussed. The test environment offers a flexible system that is capable of reacting to the dynamic changes of process demands. Physical network performance can be predicted by analysing the historical data in the background on a network simulator or virtual network. Finally, this thesis offers a conclusion with an indication of our future research work.