Abstract: The conventional research on vibration comfort relies on manually filtered data to address the issue of non-monotonic and nonlinear relationships between vibration intensity and subjective sensation caused by subjective ambiguity. This method suffers from several difficulties in determining exclusion criteria, failing in large-sample studies and, in capturing the characteristics of human subjective sensation effectively. This study leverages a multi-source heterogeneous big dataset on vibration comfort and quantifies the chaos of data using information entropy. Based on the C4.5 algorithm, the Maximum Information Gain (MIG) method is proposed, which is suitable for big data analysis. This method selects vibration thresholds that maximize the information gain after data grouping. The results are compared with traditional mean-based methods and K-Means clustering method, and numerical experiments are conducted for validation. The results demonstrate that the MIG method is more suitable than previous thresholding methods for big data scenarios with disorder.