The number of cells in the class is fewer than the minimum number to obtain valid statistics from. In a supervised classification, you identify classes by creating polygons, called training sets, that identify specific features. From the cells contained within the training set, statistics are calculated defining each class, which are stored in a signature file. There needs to be enough cells in a class to obtain valid statistics.
Add cells to the class by expanding the size of the polygon in the training set or adding other polygons that identify the same class.