Feature Selection Algorithms for High-dimensional Unbalanced Medical Data

High-dimensional feature selection is a difficult issue in medical field, while the class imbalance problem existing in medical data can also seriously affect the classification performance of algorithms. To address the joint problem of high-dimensional feature selection and class imbalance, a new objective function and the SMOTE imbalance approach are performed on the Simple, Fast and Efficient (SFE) based high-dimensional feature selection algorithm, as well as incorporating the Metropolis criterion in the SFE algorithm to help the algorithm jump out of the local optimal solution. Finally, a feature selection algorithm based on high-dimensional unbalanced medical data (GOSFE) is proposed. Experiments are conducted on six public datasets, and the results show that compared with the SFE algorithm, GOSFE improves the G-means and Fl metrics by 6.75% and 2.53% on the SMK_CAN_187 dataset, and the G-means and Fl metrics by5.95^%and 3.42% on the Leukemia dataset, respectively. Meanwhile, the experiments show that the GOSFE algorithm can quickly search the subset of features with the highest classification accuracy, reduce the redundancy of high-dimensional data, and has a good potential for improving the high-dimensional imbalance problem of medical data.