Feature selection is one of the important techniques of machine learning and data mining. However, with the continuous growth of data scale and application scenarios, the traditional feature selection methods have some defects. In this paper, a Feature Selection method based on Improved Genetic Algorithm (FS-IGA) is proposed. Firstly, the fitness function is improved by introducing the separability criterion based on the distance between classes. Secondly, the decision tree is introduced as the classifier in the genetic algorithm to optimize the performance of the algorithm. Finally, experimental results show that our proposed method can effectively improve the accuracy and adaptability of feature selection. The method presented in this paper boasts of a unique advantage in that it transcends reliance on any specific model framework, thereby ensuring broader applicability. As such, this paper provides a more profound reference value for the research on feature selection in the field of machine learning and data mining.