Efficient Neighborhood Function and Learning Rate of Self-Organizing Map (SOM) for Cell Towers Traffic Clustering

Abstract

The self-organizing map (SOM) neural network is based on unsupervised learning, and has found variety of applications. It is necessary to adjust the SOM parameters before starting learning process to ensure the best results. In this research, three types of data represent high and low traffic of specific cell tower with subscriber positions distribution in central of Iraq are investigated by self-organizing map (SOM). SOM functions and parameters influence its final results. Hence, several iteration of experiments are performed to test and analyze Bubble, Gaussian and Catgass neighborhood functions with three learning rates (linear, inverse of time and power series) and they were evaluated based on the quantization error. The experiments results show that Bubble function with linear learning rate gives the best result for clustering cell tower traffic.

An Unsupervised Anomaly Detection Based on Self-Organizing Map for the Oil and Gas Sector

“…Thus, a more homogeneous output that preserves as much as possible the initial topological composition returns. There are several neighbourhood functions in the literature, such as the Bubble Equation ( 9), Gaussian Equation (10), Cutgass Equation (11), and Epanechikov Equation ( 12) [43].…”