XAI-Driven Deep Learning Framework for Classifying Indian Music Genres

In the digital era, vast online music repositories demand efficient classification systems to organize and recommend tracks effectively. Traditional rule-based and statistical methods often fail to capture the intricate tonal and rhythmic patterns, especially in diverse forms of Indian music. This project presents a Deep Learning-based Music Genre Classification (MGC) system enhanced with Explainable Artificial Intelligence (XAI) techniques. The system employs Mel-Frequency Cepstral Coefficients (MFCCs) to extract meaningful frequencydomain features from audio signals, which are then processed by an Artificial Neural Network (ANN) for genre prediction. Five major Indian genres—Bollypop, Carnatic, Ghazal, Semi- Classical, and Sufi—are classified using this model. To enhance interpretability, LIME, SHAP, and Permutation Importance methods are integrated, providing insights into the model’s decision-making process and feature contributions. Experimental results demonstrate that the proposed ANN model achieves 99% training accuracy and 81% validation accuracy, outperforming CNN, RNN, LSTM, and hybrid architectures used for comparison. Beyond its technical significance, this system contributes to the preservation and accessibility of Indian musical heritage, supporting applications in music recommendation, digital archiving, and intelligent content organization.