Scalable IoT Analytics with Federated Learning: A Convex Optimization Approach Using Machine Learning Algorithms

Abstract

The increasing Internet of Things (IoT) network of connected devices generates enormous amounts of data that may be evaluated and used to inform decisions. The variability and diffusion of IoT data provide significant challenges for machine learning models, which typically require a lot of data to be taught. By leveraging the data they have locally collected, several devices can collectively create a global model using federated learning, a new approach to machine learning, without sharing the raw data with a central server. In this study, we present a federated learning technique for scalable IoT analytics based on the stochastic gradient descent (SGD) algorithm. In order to collectively create a global model for predicting energy demand, our technique makes use of a number of IoT devices, including smart meters. The global model can be divided into smaller components that can be trained concurrently on many devices using a distributed technique based on SGD, which we also recommend. Our research demonstrates that our method is more precise and scalable than traditional centralized learning algorithms using a real-world dataset of smart meter readings. Our method also provides stronger privacy safeguards because the raw data is stored locally on the devices rather than being shared with a centralized server. Our recommended methodology offers a novel strategy for resolving IoT analytics’ challenges and exemplifies the promise of federated learning for doing so in a distributed and private manner.