ABSTRACT
With the rise of Internet of Things (IoT) devices, there is a growing demand for artificial intelligence algorithms appropriate for ultra-minimal IoT settings. This called for a study to develop machine learning and deep learning models such as Random Forest, Logistic Regression, and Convolutional Neural Networks which must possess low latency, small model sizes, and low power consumption, all being crucial for the deployment of IoT platforms. Optimization techniques like quantization and pruning were aimed at improving efficiency, yet targeting achieving accuracy under strict conditions implied by an IoT framework, the AI models were trained using the PAMAP2 dataset. They are converted into TensorFlow Lite, TVM, UTensor and other ML Light Weight Frameworks which permitted their implementation making use of edge computing architectures after exhaustive assessment using the Edge Impulse platform which specializes in applying ML model over marginal hardware yielding significant practicality test results virtual confirmation distinct real-world situations; this is brought on by precision, latency, and CPU utility assessment, so a lot useful information concerning an efficient choice of algorithms in a variety of test scenarios that remain primarily centred on a minimal level application, the result expected to somehow assist scholars, professionals, and many others in making smart choices, hence having a creep effect across disciplines specifically toward computation edges, IoT device deployments pan out in real cases of necessity-to-tailored solutions optimally in resonance, brings wide appeal relevance, outreach, providing resource utilization at par optimum performance and is quite insightful, especially where resources scarcity crops up.
