I used Gaussian Mixture Model (GMM) for classification and I want to improve its performance. In my experiences with NN or other classifier algorithm, we can adjust the parameters to improve the performance. What about the GMM?
hi
you can improve the GMM performance by applying some optimizations in it.
ALso you can apply preprocessing to your data.
Thank you.
Best regards
Dr.Indrajit Mandal
Ka-Chun Wong, I don't understand about regularization. Would you explain it? I forgot to write down more detail about my case. I want to recognize emotion from biosignals. There are 3 classes. So the number of Gaussian distribution is fixed to 3.
Indrajit Mandal, would you explain about optimization in GMM?
For each class, you should have a separate GMM (multivariate Gaussian distribution). The performance is to certain extent related to number mixture components that you use. You can try increasing the number of mixture components. For a 3-class problem, conventional GMM is expected to give close to 100%. If you try to identify emotion from speech signal, try increasing the duration of the speech signal during training and testing.
Nagarajan, I use biosignal from the Mahnob-HCI database. I have 7 classes and train 7 GMM distributions. The final result, only 3 out of 7 GMM distribution used by the system. I think when I increase the number of GMM dist., I will get the same result. I am thinking to regularize my data but I need to learn about regularization technique. Do you know?
- Badges
- Science topic