Feature selection is an effective method for dimensionality reduction in machine learning and data mining. However, it is challenging to select the optimal feature subset with smaller size and higher classification accuracy from high-dimensional data. In this paper, a new approach for feature selection using multi-objective optimization algorithm with improved neighborhood rough sets is proposed. Firstly, the improved neighborhood positive region considering the classification information in the boundary domain is presented to measure the importance of feature more accurately. Then, two optimization objectives are designed to evaluate the quality of the candidate feature subsets. The non-dominated sorting operator and the crowding distance operator are employed to obtain the optimal solution sets. Finally, we utilize the feature kernel to study the relationship between the solutions in the same Pareto front. The performance of the proposed algorithm is examined on ten benchmark data sets and the results are compared with state-of-the-art algorithms to verify the validity. Experimental results show that the proposed algorithm can obtain the high-quality tradeoff between feature subset size and classification accuracy.