Journal of Intelligent Computing Jianli Ding, Liyang Fu 9 3 2018 http://www.dline.info/jic/fulltext/v9n3/jicv9n3_1.pdf As an important pre-processing method in machine learning, feature selection eliminates data redundancy and reduces feature dimensions and computational time complexity effectively. In order to further reduce the number of iterations of the feature selection algorithm and improve the classification accuracy, a new hybrid feature selection algorithm is proposed, which combines the filter algorithm based on information gain and the wrapper algorithm based on Sequential Forward Floating Search (SFFS) and Decision Tree (DT). The optimal candidate feature subset is quickly found by ranking the features using information gain. In order to avoid the nesting effect of features, SFFS algorithm is used to reduce feature dimensions of the optimal candidate feature subset. The experiments show that the maximum ratio of the number of reduced features and the number of initial features is 92.86%. Compared with other feature selection algorithms, the maximum decline of the number of iterations is about 67.8%, and the maximum increase of the classification accuracy is about 10.5%. The results prove that the hybrid algorithm possesses higher computational efficiency and classification accuracy.