A new method called Modified Kernel-Based Fisher Discriminant Analysis (MKFDA) has been developed to improve classification accuracy in data analysis. Unlike the traditional method that only gives one discriminant vector, MKFDA can provide multiple discriminant vectors, enhancing its ability to classify data accurately. By re-estimating the scatter degree between different classes in the data, MKFDA can better separate and classify information. Testing on the IRIS dataset has shown that MKFDA is effective in improving classification performance.