GT INDEX - MEASURE OF THE SUCCESS OF PREDICTION

In order to evaluate the results of a remote sensing study, ground truth data often are collected and a cross-classification table is prepared. However, the commonly used Percent Correct Classification criterion may be misleading. An operationally meaningful measure of success of prediction is proposed. An example using corn blight data demonstrates the value of this new index and casts some doubts on earlier conclusions based on the percent correct classification criterion. © 1979.