Figure 3
From: Using self-organizing maps to develop ambient air quality classifications: a time series example
Measures used to aid selection of number of day types for SOM classification system. Panel (a) two-dimensional representation of multipollutant data created using multidimensional scaling. Panel (b) Calinski-Harabasz clustering index values for each class number tested. (Higher values are better.) Panel (c) resulting mean (± range) adjusted R2 values from regression models fit to each pollutant using the SOM classification as the predictor for each class number.