The classic support vector machine(SVM) can not efficiently exploit the local information of data points, which
is useful for pattern recognition. Therefore, a so-called local learning based support vector machine is presented to address those problems mentioned above, which makes full use of the local information such as intra-locality scatter and inter-locality scatter of the data sets to search an optimal decision function by minimizing the intra-locality scatter and simultaneously maximizing the inter-locality scatter. Meanwhile, the proposed method adopts geodesic distance metric to measure the distance between data, which can reflect the true local geometry of data space. In addition, an additional parameter ?? is introduced to control both the super bound on the fraction of margin errors and the lower bound on the fraction of support vectors, thus improving the generalization capacity of the proposed method. Finally, extensive experiments show the effectiveness of the proposed method on the artificial and real world problems.