When attempting to perform an unsupervised classification, the process does not complete due to the error “Unable to invert eigenvectors”.
The error message “Unable to invert eigenvectors” is basically saying that the classification process is unable to invert the covariance matrix. This could mean that the input data is too homogenous and there are too many similarities within the layers. For example, if two or more layers in an image are the same or if an area of the image has the same pixel value in each band it may not be possible to invert the covariance matrix or find a group of unique pixels within the data.
The Unsupervised Classification tool will not let you choose which layers to include in the classification, so you must remove the unwanted layer(s) prior to performing the classification. You can use the Subset tool to remove a layer(s) from your image. The Subset tool can be accessed by going to Raster tab > Geometry group > Subset & Chip menu > Create Subset Image.