[Week #7 — Guess The Genre]
We’re Atalay Gürel, Muhammed Fidan & Ozan Postacı. We are students from Hacettepe University Computer Engineering Department and this is the last article of our Machine Learning Course Project about Music Genre Classification.
MLP
In previous weeks we implemented MLP with 8 genres and accuracy was around %45. After removing Folk genre it increased to %54 accuracy. We tried some activation function(tanh, ReLu) but we got better results in ReLu activation function in MLP model training.
The model consists of two hidden layers. There are 300 nodes in the first hidden layer and 32 nodes in the second hidden layer. We observed %54 accuracy in the test data after 30 epochs.
Conclusion
In our project, we trained these 6 different models and tried to compare their results.
The same style of instruments is commonly used in some music genres, mostly guitar. This caused the models to confuse while learning, so the results were in bad condition with every model. That’s why we decided to remove the folk music genre from the dataset to reduce confusion during model training.
After removing the folk genre from the dataset, we observed a nearly 10% improvement in success in each model. We found an 8% difference between our highest and lowest models.
The results in general were close to each other.
The method with the lowest accuracy value was KNN. Logistic Regression, Deep Neural Network and Random Forest results were similar. We can say that the best models were SVM and MLP.
Stay tuned with your music and let us guess the genre ♫
