Gradient visualization of grouped component planes on the SOM lattice

The Self-Organizing Map has been successfully applied in numerous industrial applications. An important task in data analysis is finding and visualizing multiple dependencies in data. In this paper, we propose a method for visualizing the Self-Organizing Map by decomposing the feature dimensions into groups with high correlation or selections by domain experts. Using Gradient Visualization we plot a vector field for each of these groups on top of the map lattice, with arrows pointing towards the nearest cluster center. We provide a real-world example from the domain of petroleum engineering and point out our technique’s usefulness in understanding mutual dependencies hidden in the data.