An optimized method for classifying data called kernel Fisher discriminant has been developed. This method is more efficient in sorting data into categories. The researchers created algorithms for sorting data into two groups or multiple groups. They found that only using a portion of the data, called "significant nodes," is needed to classify new data accurately. By selecting these "significant nodes" using a special algorithm, the method is effective and efficient in classifying data, as shown in experiments with benchmarks and face image databases.