Tumor gene expression data feature selection method based on locally linear embedding algorithm

The invention provides a tumor gene expression data feature selection method based on the locally linear embedding algorithm. According to the method, a neighbourhood is calculated according to class information of tumor gene expression data, and for better utilizing the class information, a new distance expression mode is redefined, wherein I is the distance between identical label samples, and II is the distance between different label samples; reconstruction weights of samples in a class and reconstruction weights of samples outside the class are calculated respectively; rules are distinguished; feature evaluation is conducted on a function. The tumor gene expression data feature selection method has the advantages of the LLE Score algorithm that high-dimensional neighbourhood information can be retained in a low-dimensional structure, label information can be well used, and meanwhile the calculation cost is low. According to the tumor gene expression data feature selection method, gene data can be classif