To address feature redundancy and over-fitting resulting from excessive data volume and high feature dimensions in speech communication interference assessment, this study employs dimensionality reduction techniques. Firstly, a simulated dataset of disrupted speech is created to mimic intricate speech environments. Subsequently, the disrupted speech dataset is manually and subjectively annotated. Next, machine learning algorithms are employed to categorize the feature metrics of disrupted speech. Finally, the original feature metrics undergo dimensionality reduction, and the processed data is compared to the pre-reduction data. Experimental outcomes indicate enhancements in the results post-processing with the three dimensionality reduction methods. Particularly, Linear Discriminant Analysis (LDA) processing notably enhances accuracy. Moreover, Multidimensional Scaling (MDS) and Principal Component Analysis (PCA) also demonstrate significant improvements, especially in Mel frequency cepstral coefficients (MFCC) and wavelet statistical characteristics. This validates the effectiveness of the proposed approach.