Interpretable Content-Based Music Genre Classification Utilizing a Modified Artificial Immune System with Binary Similarity Matching

Abstract

This study investigates the application of a modified Negative Selection Algorithm (NSA), derived from principles of the human immune system, to enhance music genre classification. NSA’s threshold-based similarity matching mechanism plays a pivotal role in distinguishing genre-specific patterns, yet its optimization remains underexplored in music information retrieval. The proposed framework integrates censoring and monitoring modules to refine classification boundaries and reduce misclassification rates. It focuses on three core musical attributes: timbre, rhythm, and pitch, extracted from vocal, melodic, and instrumental elements. These features undergo systematic extraction, selection, and categorization to improve genre identification and labelling accuracy. Experimental results across diverse threshold settings demonstrate that the modified NSA achieves competitive performance compared to conventional classification models. The findings highlight NSA’s adaptability and robustness in handling genre variability, especially in cross-domain music datasets. Beyond technical contributions, this study emphasizes the importance of understanding musical features that define genre identity. By offering a biologically inspired, threshold-sensitive model, the research contributes to the development of intelligent, interpretable systems for multimedia classification. The approach supports more accurate music categorization, which has implications for recommendation systems, digital archiving, and cross-cultural music analysis. This work bridges computational intelligence and music analysis, offering a novel perspective on immune-inspired learning for content classification. It reinforces the potential of NSA as a practical and scalable tool for genre recognition in diverse musical contexts.