<p>Feature Selection (FS) is an essential preprocessing step for multi-label (ML) data, which is typically high-dimensional and complex and can be affected by the curse of dimensionality. Multi-label feature selection (MLFS) approaches are widely used to increase their performance while meeting a variety of FS goals. In addition, label types for FS are also possible, such as text, audio, and video files. Several techniques were developed to provide an efficient MLPS approach, but fewer limitations were obtained, such as low performance, high error rate and time complexity. To overcome these kinds of issues, this paper proposes a lightweight FS approach using the Preference Selection Index (PSI), a Multi-Objective Decision-Making (MODM) method. The proposed PSI-based multi-label FS method (PSI-MFS) considers three conflicting objectives: minimizing Feature-Feature Redundancy, maximizing Feature-Label Relevance, and maximizing Feature-Label Interaction. Then, a decision matrix based on these objectives should be formulated, in which the characteristics are considered alternatives, and goals are considered criteria. PSI-MFS then calculates the PSI value to rank the features. The time complexity of PSI-MFS is low, which provides better performance than existing models for ML classification. The performance of PSI-MFS is evaluated on 10 benchmark datasets and compared with 11 baseline FS methods. The results are statistically tested using Friedman's test, which shows that PSI-MFS is statistically significant. It is also observed that PSI-MFS requires less execution time and outperforms other well-known methods 80% of the time. This technique achieves more than 68% accuracy for 10 datasets and reduces the loss rate below 0.200%.</p>