One of the modern era's emerging technological breakthroughs is the Internet of Things (IoT), which has drawn significant interest from businesses and academia. IoT has made objects (things) smart with sensing capability and linking things to the Internet. It has immense potential, and there is a lot of buzz around it. It also promises to raise people's quality of living by introducing cutting-edge technology and integrating it into daily life. Hence, businesses and city administrations are adopting IoT-based solutions to deliver effective services. Because IoT is critical in advancing social and infrastructure growth for information and data-oriented services at every level, promoting economic development.

However, the rapid growth of IoT has come with heightened security threats and privacy issues due to resource constraints of IoT devices and a lack of intelligence. As a result, academia and the industry have explored various approaches to IoT security, with recent studies showing significant interest in adopting machine learning (ML) models for anomalies and intrusion detection systems to improve IoT security.

In the area of security and privacy in IoT, various studies have explored the adoption of Artificial intelligence to make IoT devices intelligent for real-time vulnerability and attack detection. Thus to gain a deeper understanding of how ML can be deployed at the edge devices, this study aims to investigate how industry players can leverage ML to improve security in IoT.

Based on the quantitative approach, data sets generated by u-blox, Malmö testbeds, and UNSW-NB 15 network data sets are used to build, train, and evaluate ML models. The data sets were subjected to an exploratory analysis, which provided participants with practical information on the model's performance, best features, and compute time and resource needs. Similarly, the qualitative approach involved a mini-focus group and individual interviews with security, product development, embedded systems developments, and machine learning experts from u-blox. The empirical data gathered from interviews and mini-focus group meetings were analyzed using thematic analysis to produce four themes and results and analysis of ML models to complete the empirical findings. The study subjects were then compared to the analysis, outcomes, themes, and ideas from the literature review as well socio-technical perspective based on the affordance theory to embedded ML was considered.

The findings of this master's thesis demonstrate that ML models can improve real-time anomaly and intrusion detection in IoT by improving intelligence at the edge devices. More so, ML deployment at the edge device requires a system development approach because of the resource constraints of IoT devices. Furthermore, ML deployments require integrating embedded software development, ML, and business skills. Also, IoT devices are resource constraints, cheap, low energy powered, and have inadequate computing power; hence require TinyML for ML model deployment. Finally, the study suggests an affordances-based socio-technical and system development approach to overcome the hardware, software, and system stack hurdles and integrate ML models for behavioral modeling and IDS for secure, reliable, simple-to-use, and transparent IoT edge devices.