Explainable AutoML-Driven AI-Assisted Traditional Music Composition and Genre Classification via MFCC-Based Acoustic Feature Learning

Abstract Accurately classifying music genres is important for music information retrieval, recommendation systems, and traditional music heritage conservation. A high-accuracy interpretable music genre classification system was designed. This study proposes an artificial intelligence-aided approach that combines MFCC-based acoustic feature learning, spectral and harmonic descriptors, explainable automated machine learning (AutoML), optimised ensemble learning, and SHAP-based explainability for traditional music genre classification and composition. It was implemented based on 5-fold cross validation with bagging, stacking, and weighted ensemble learning on the considered music genre dataset known as the GTZAN. Furthermore, genre-specific prototypes of acoustics are learned to support music generation based on AI assistance, which can provide feature targets that are representative of the music style for generating music that is consistent with the music style. The experimental results revealed that the WeightedEnsemble_L3 ensemble model achieved a maximum accuracy of 94.59%, whereas the individual classifiers were classified at a lower rate. SHAP makes it easier to identify the most important acoustic characteristics for the model, increasing its transparency and interpretability. Moreover, knowledge-based acoustic prototypes allow the acousability of the generated music to be quantified in comparison to the given music-making genres and transfer this context into AI-assisted music generation. The suggested framework offers a genuine, comprehensible, and viable solution for traditional music recognition and supports culturally motivated AI aids to music production.