I am trying to select a right neural network structure for my data. I have similar input data, however, the number of classes are different for various cases. Moreover, more than one labels can be accepted as a true label. Can you suggest the best neural network structure that fits my data type?
For Multi label classification you can use deep neural network model with sigmoid function.
You have mentioned more than one label is accepted as true label for such cases you can go for ensemble based learning models.
Priyatharshini Rajaram Do I need to consider convolutional neural networks? or graph neural networks? Do you think they can be helpful?
Can u share your problem statement so that i can suggest better model
In multi-label classification, one data sample can belong to multiple classes(labels). Where in multi-class classification, one data sample can belong to only one class. So, i recommend you to read this tutorial: Multi-Label Image Classification with Neural Network
https://towardsdatascience.com/multi-label-image-classification-with-neural-network-keras-ddc1ab1afede
You could try Auto Model from RapidmIner and see which model gives you the best results.
According to you requirement you go for ensemble techniques like random forest
I suggest you follow for better understanding Multi-Class Classification Analysis: Article A Survey of Multi-Label Topic Models
https://link.springer.com/content/pdf/10.1007/s10994-008-5064-8.pdf
https://pdfs.semanticscholar.org/6b56/91db1e3a79af5e3c136d2dd322016a687a0b.pdf
Fuzzy or neuro-fuzzy classifiers generate a probability value for each sample belonging to all classes
For a problem with more than 2 classes. A multiclass neural network model can solve such problems. However the choice of the model depends on the problem.
You can go for convolutional neural network architecture. You can use relu activation in each layer of convolution. In the fully connected layer use softmax function which is suitable for multiclass classification
- Badges
- Science topic