Regional folk music has a deep historical aspect and reflects an excellent cultural heritage. This paper examines how machine learning can be used to classify regional Indian folk music. The sample includes 593 songs in six states, Maharashtra, Gujarat, Punjab, Assam, Bengal and Uttar Pradesh which is equivalent of 2,965 audio clips. Both time and frequency domain features, including Mel-frequency cepstral coefficients (MFCCs), onset strength, Mel-spectrogram, spectral bandwidth, spectral contrast, tonnetz, and zero-crossing rate were extracted and utilized in features classification between Light Gradient Boosting Machine (LightGBM), Random Forest (RF), K-Nearest Neighbors (kNN), as well as Support Vector Machine (SVM) class learning algorithms. Combining short-term spectral features showed a great improvement in classifying images as opposed to using individual feature sets. Out of the tested models, LightGBM had the highest accuracy of 82.09 and F1 score of 0.82, which is better than another model kNN, SVM, and RF. In addition to its technical outputs, this work presents a useful prospect in cultural preservation and digital archiving in being the means to allow the systematic identification, organization, and recovery of regional folk traditions, facilitating the conservation of heritage in the digital world.