Efficient Kernel Design of Support Vector Machine for IoT Networks

A high-level goal of 6G cellular networks will be achieved by enhancing the current 5G technologies as well as adopting new techniques such as artificial intelligent (AI). The AI will be a new blood to improve 6G cellular networks. The AI techniques enable us to find the hidden patterns from cellular network data and optimize cellular networks. The Internet-of-Things (IoT) will be a key application of 6G systems and play an important role in 6G ecosystem by creating many new business models. As 6G systems have a higher requirement of networks, the IoT networks in 6G should be designed more efficiently. We expect to improve the performance of IoT networks significantly by adopting AI techniques. The support vector machines (SVMs) as one of key machine learning techniques solved many real-world problems in many different areas such as face recognition, imagine processing, and so on. Since the kernel method as a key component of SVMs is computationally cheaper than the explicit computation, it is widely used when implementing SVMs. Selection of a kernel function directly affects to the performance of SVMs. In this paper, we investigate kernel function design and propose a new design method using a linear combination and K-fold cross validation. We show that the proposed scheme reduces approximately 30% misclassification error rather than conventional single kernel.