Multi-label Classification with ordinal value

I just faced a problem that I never experienced. My goal is to predict the number of times they will eat food in cafeteria from the rating (1-5) of each menu there are total of 34 menus. When I implemented logistic regression or feedforward network to tackle it, it did not work as both the accuracy and loss did not decrease.

Loss function: Cross entropy

for logistic regression, I have 3 hidden layer.

One think I am suspecting is that I have too less data, which is around 200 points. This is kind of prototype but could you guys maybe give some suggestions on tacking this issue.

Hello Jae,

I have similar task too, right now im just doing simple multiple output regression with MSE as a loss function. How did your research going till now? Could you share what you’ve done in this kind of task?


I think the discussion in Create a Variation of Cross Entropy Loss with Per Pair of Classes Weights might be useful.