Distributed data storage using decision tree models and support vector machines in the Internet of Things

Abstract

The rapid development of IoT technologies generates a considerable amount of diverse and distributed data, mostly real-time and sensitive. Due to the diversity of data types (text, image, video) and geographical dispersion, efficient management becomes essential for maintaining performance and ensuring speedy responses to users. Traditional data storage methods are unfit for dynamic IoT environments, due to their lack of scalability, energy efficiency, and bandwidth. Recent research indicates that machine learning might offer enhanced security with reduced latency and improved energy efficiency. However, most of these techniques are complex and resource-intensive, hence inappropriate for resource-constrained IoT devices. While various developments have been made in this regard, a holistic approach that not only forecasts the requirements for data replication but also selects the most optimized storage nodes remains an unmet challenge. The presented paper offers a hybridized approach by incorporating Decision Trees and SVM, which manage data optimally with higher speeds and reduced computational costs. Simulation results indicate that this method can reduce access latency by up to 22.2–41.6 %, increase accuracy by 5–12.3 %, and improve resource utilization efficiency by 7.7–15.3 %.